Home > Blog
Cassantecs Blog about Prognostics
Prognosticating is often associated with clairvoyants and crystal balls that use magic to predict future events. Yes, we also predict future events. Yet, the method we use for that could not be more of the opposite from the above. Also, we heard the question whether Prognostics is the same as predictive analytics more than once. In short, we often encounter ambiguity about the word Prognostics.
Be it in the US, Germany, Switzerland or East Asia - Prognostics is a novel tool for maintenance management anywhere and requires explanation. We introduce the Cassantec blog to make the concept of Prognostics better understood and write about actual projects to make it more tangible. Of course we will also inform you about the latest product updates and relevant company news.
Blog Contact Us
– 12 January – written by Moritz von Plate, CEO of Cassantec
Podcast - Prognostic Analytics vs Predictive Analytics in IoT
Listen to our IoT Business Podcast - Prognostic Analytics vs Predictive Analytics in IoT Click here
– 10 November – written by Moritz von Plate, CEO of Cassantec
Catch of the Week
– 22 September – written by Moritz von Plate, CEO of Cassantec
Captain Kirk and Mr. Spock are perfect examples for how the combination of intuition and logic yields superior results. Captain Kirk would always consult Mr. Spock, counting on his impeccable fact-based reasoning. He would way the arguments and then make the best possible decision.
In industrial asset management, Cassantec takes the role of Mr. Spock. The transparent, impartial, repeatable, logical and fact-based prognoses are input, if not foundation, for the decisions of asset and reliability managers, the Captain Kirks of industrial operations.
Or, like a customer of mine once said: “Let’s view Cassantec like another person in the room to help with making decisions.”
So, if you are still lacking a Mr. Spock at your side, please approach us!
And let’s hope that your Mr. Spock will never have to say: “Your logic was impeccable, Captain. We are in grave danger.”
– 01 September – written by Moritz von Plate, CEO of Cassantec
What can weather proverbs (“Bauernregeln”) teach us about heuristics and data analytics?
Probably the best known weather proverb (“Bauernregel”) in all of Germany is the “Siebenschläfer”. It goes:
"Regnet's am Siebenschläfertag, so regnet's sieben Wochen danach"
– if on June 27 there is rain, there will be rain for the following seven weeks.
This heuristic is based on century-old weather observations. And indeed, just around that time of the year, the jet stream settles geographically. And depending on whether it settles further north or south, it has an impact on the summer’s weather pattern.
But what hardly anybody in Germany knows is that the rule’s accuracy varies strongly by region:
• In Munich it has an accuracy of around 80%
• Further north in Berlin, the accuracy is in the high 60s
• In Hamburg, the rule does not work
And the proverb is from the south of Germany. Hence, it is hardly surprising that this is where it actually works. But ask a person from northern Germany (like my mother), and she will happily quote the “Siebenschläfer”-rule, not knowing that it is not applicable to her local weather.
Luckily, we nowadays all have 2 or 3 weather apps on our smartphone that can actually fill the void and provide much more accurate forecasts based on data analytics. So, we don’t have to rely on proverbs and heuristics anymore, of which we don’t know whether they actually apply.
Transfer that to industrial applications:
Many a maintenance strategy still works based on heuristics. For example: “Based on decades of experience, the turbine needs maintenance after 3 years of operation. The scope of required repairs increases dramatically, if we prolong that to 4 years.”
But is that heuristic experience still applicable after the dramatic changes of the last years? Don’t renewables lead to new operating regimes, new loads on the equipment and changed aging processes?
Doesn’t the increased cost pressure lead to a new cost optimum?
Nobody really knows.
I recommend: rather than relying on possibly false experience, use data analytics and be on the safe side!
Back to the Bauernregeln, this one is always right:
“Kräht der Hahn auf dem Mist, ändert sich’s Wetter oder es bleibt wie es ist!“
- “If the cock crows on the dot, it will rain or it may not.”
– 08 July – written by Moritz von Plate, CEO of Cassantec
The T-model of data analytics: how come that “predictive analytics” doesn’t predict the future?
In data analytics, both depth and breadth are desirable but can’t be had from the same tools.
Therefore, we follow the T-model of data analytics. While there are many companies focusing on depth (the vertical bar in the “T”), Cassantec delivers the missing breadth (the horizontal bar in the “T”).
What do I mean by that when it comes to Predictive Maintenance?
• Depth provides diagnostic insight into the current equipment condition, going towards root
cause analysis to answer questions like why, where and how.
• Breadth generates prognostic foresight into the future equipment condition. It addresses
the when-questions. Here, diagnostic depth is complemented by prognostic breadth.
One leading provider of depth recently wrote: “XXX is often classified in the equipment reliability world as a “predictive analytic” product. However similar to other technologies in this space, it doesn’t “predict” the future, it basically finds deterioration in equipment performance well before it’s obvious, with very high confidence, and across many, many types of equipment. […] What we’ve seen over the years, is that while XXX does find the kinds of issues that could lead to catastrophic failure, it also finds hundreds of other problems, many that no Failure Mode Effects Analysis (FMEA) or standard maintenance program would anticipate.”
- This is a great example of analytical depth, which is somewhat blurred by the term “predictive”.
Such outstanding depth must be complemented by Cassantec’s breadth in order for asset managers to have insight as well as foresight, both of which are required to optimally deploy their equipment. We are unrivalled in calculating explicit prognostic horizons over weeks, months, and in cases years (for any given time in the future we calculate the risk of malfunctions occurring).
So, adopting the T-model wasn’t only central for manufacturing 100 years ago. It is also paramount to lifting today’s asset management to the next level.
– 10 June – written by Moritz von Plate, CEO of Cassantec
"We have similar prognostic tools ourselves" - really?
Regularly, I hear from potential customers that they or their current services provider have tools that provide prognostic information. Really?
I suggest you ask the following probing questions:
• What is the prognostic horizon of these tools?
[if it is not weeks and months, it is not prognostic!]
• How explicit is the prognosis?
[if it is just an early warning (“beware, malfunction x is about to happen”) rather than actual
information on when to expect the malfunction, it is a better condition monitoring system,
but not prognostic!]
• Does it cover all relevant components and malfunctions drawing on all available process and
condition data, e.g. also lubricant analyses?
[if it covers just a subset of components, malfunctions and/or data, it is nice for the covered
portion, but not capable of delivering the full benefit of prognostics!]
If, on the other hand, the above questions can be answered satisfactorily, congratulations!
....and please let me know, because I would love to get a first-hand impression on that competitor of ours....
– 20 May – written by Moritz von Plate, CEO of Cassantec
The Use Cases of Prognostics
Somebody recently asked me: “why would I want to know when my equipment will develop a malfunction?” Well, until that moment, I thought it pretty obvious that having a peak preview into the future is rather helpful in many ways.
That question triggered me to make a list of the many use cases for knowing the future in industrial and equipment operations. Here it is:
Use case category “Maintenance and Repair”:
1. Long-term scheduling of maintenance
(Optimize scheduling and scoping of outages to secure high availability with minimum budget)
2. Short-term preparation of reactive maintenance
(Enhance detection of approaching inefficiencies, malfunctions and failures through early warning tool with explicit quantitative risk information)
3. Maintenance staff planning and allocation
(Optimize staff planning and allocation with maximum impact on availability and minimum fire-fighting potential)
Use case category “Operations”:
4. Production planning
(Adjust operations and production plan to the future availability profile, considering known risks and related contingencies)
5. Mission forecasts
(Select assets with sufficient expected availability and low downtime risk for field projects, e.g. in upstream oil & gas, and minimize the set needed to secure committed availability)
Use case category “Finance”:
6. Generation availability forecast for power trading
(Ensure assets availability at peak energy prices, schedule maintenance intervention at lower prices, and consider the quantitative generation risk – along with the known quantitative market risk – in a joint commercial model)
7. Planning budget(s) and total cost of ownership (TCO)
(Substantially improve longer-term forecasts of repair and replacement costs at all levels for the coming fiscal years by considering the current asset conditions beyond cost histories)
8. Optimized insurance policy and costs
(Negotiate lower insurance premiums by showing quantitative risk profiles and future availability forecasts based on condition and process data not (yet) considered by the insurer)
Use case category “Procurement”:
9. Optimized parts and service procurement
(Optimize procurement timing and scoping of replacement parts and services, inventory management, consumables)
Use case category “General Management”:
10. Management reporting
(Standardize top-level management perspective via prognostic report fleet view, both as quantitative decision basis and as qualitative visual communication device)
11. Risk reporting
(Operationalize risk valuation standards, set risk acceptance/rejection criteria, and distinguish likelihood and impact dimension of risk)
12. Knowledge management
(Train maintenance workforce and reliability managers through prognostic solution:
malfunction mode definitions and dependencies, condition parameter specification and
dependencies, condition parameter specification and correlation, risk profiles over time, etc.)
(Display risk profiles, recognize risk-impact of different operations strategies, and
manage risk profiles over time, and set risk acceptance/rejection criteria)
(Align internal learning and process adjustment process with automated learning
process of solution configuration across the entire asset fleet)
13. Health, Safety & Environment (HSE) reporting
(Improve safety by minimizing risk, automate compliance reporting with respect to regulatory requirements and norms)
Use case category “Life Cycle Management” (LCM):
14. Replacement and retrofit planning
(Achieve RUL-optimal refurbishment, considering asset criticality, availability, reliability and malfunction risks)
Use case category “Product Development”:
15. Design and development of industrial assets
(Create feedback loops from operational condition & process data via remaining useful life (RUL) distribution back to original asset design)
16. Retrofit and/or replacement of sensors
(Upgrade data acquisition to effectively detect and manage downtime risk and to maximize long-term availability)
Anything missing? Stay tuned for more to follow…
– 18 December – written by Cassantec
Condition-Based Malfunction Forecasts for Commercial Operations
Deployment of a Prognostic Asset Management Solution for CCI’s Power Generation Assets
Background: Castleton Commodities International (CCI) is a leading global merchant energy company, trading energy commodities and operating a variety of energy assets. These include dual-fired (oil & gas) generating units and cogeneration units in the USA, ranging from 77 MW to 600 MW and covering peak demand in metro-politan areas such as Dallas and New York City. All generating assets are fully equipped with condition monitoring and diagnostic systems, recording and archiving condition and process data for all crucial asset components.
As a power generator primarily focused on commercial operations CCI seeks full transparency of both market and generation risk, which is largely driven by the risk of unscheduled generating unit downtime. To better understand and manage generation risk on the basis of their assets’ actual conditions, CCI has introduced Cassantec’s prognostic solution across its entire fleet of generating assets, including Roseton in Figure 1. With the acquisition of additional generating assets, the solution is gradually expanding.
Objective: Objective: CCI’s objective is to actively manage the future availability of its power generating assets, in line with its ongoing commercial commitments, and to secure power supply in upcoming peak demand periods. The prognostic solution, in particular, is to provide a daily update of unscheduled future downtime risk for its generating assets. The risk of unscheduled downtime is computed at the component level, then aggregated to the unit and fleet level, and explicitly compared to power-zone-specific market prices. For units that are “in the money” at any future point in time, risk-mitigating action is taken if necessary. Such action includes preemptive yet informed component maintenance or replacement, ideally during idle periods or scheduled revision cycles.
Approach: Approach: The prognostic solution provided by Cassantec is computing future risk profiles at component, unit, and fleet level, based on condition and process data of the generating assets. This data includes cur-rent and historical parameter values for all crucial asset components, such as gas and steam turbines, power generators and transformers, boilers and HRSGs, boiler feed pumps, induced and forced draft fans. In a first step, the prognostic solution uses current and historical condition and process data to project the component’s condition into the future. In a second step, the future condition is correlated with component-specific malfunction modes, to determine the future malfunction risk for all components considered. In a third step, the malfunction risks are illustrated and aggregated in a prognostic report that offers compo-nent, unit, and fleet level views, as displayed in Figures 2-4. The generation risk reports are complemented with market power price forecasts to determine both megawatt@risk and margin@risk indices. Monitoring these indices allows adjustment of trading and hedging strategies, mitigating both market and generation risk of the power business.
Benefits: CCI is expecting benefits of the prognostic solution on three levels:
- Competitive commercial advantages at the fleet level, through the use of the prognostic reports for informed commitments in power trading and hedging
- Higher uptime records at the unit level, through the use of the prognostic reports and related down-time risk profiles for improved maintenance planning, scheduling, and scoping
- Lower asset management costs at the component level, through informed mitigation of perfor-mance flaws, inefficiencies, and latent defects and through targeted work order preparation
For reliability managers and mechanical engineers onsite, availability forecasts are a strong complement to the condition monitoring and diagnostic systems in place, consolidating data from different sources, ex-tending insights by an explicit future time dimension, and rendering standardized and conclusive reports. The forecasts may also serve as a shared planning tool in collaboration with equipment vendors, service providers, or insurers.
Next Steps: Continuous refinement of the operational solution, and extension to newly acquired generating assets.
– 24 November – written by Cassantec
Condition-Based Malfunction Forecasts for Mining Equipment
Deployment of a Prognostic Asset Management Solution at ENRC’s Frontier Mine
Background: The Eurasian Natural Resources Corporation (ENRC) operates a diversified portfolio of natural resource assets worldwide, including iron ore, copper, and cobalt mines in Africa. Due to increasing cost pressure on its mining operations, ENRC seeks to strengthen its asset management through innovative and efficient digital solutions. In early 2015, ENRC configured, tested, and deployed Cassantec’s prognostic solution at its Frontier Mine in Katanga, DRC, seen in the upper picture. In a first release, the solution scope comprised the mine’s main crusher, two cyclone pumps, and the SAG mill. In a second release, the operator is adding the ball mill, pebble crushers, and conveyors.
The prognostic solution is based on condition and process data recorded for all crucial mining assets. This data includes load, vibration, and lubricant data, recorded by Cassantec’s partner WearCheck onsite in monthly intervals, as well as vibration and temperature data taken from the operator’s online monitoring system.
Objective: ENRC is seeking an accurate, consolidated, and transparent reporting solution at the core of its mining asset management, providing substantial insight and foresight for strategic and operational decisions. Given equipment-specific maintenance plans with much flexibility around schedule and scope, the operator wishes to minimize maintenance cost and effort at constant levels of reliability and availability. While ENRC is initially using the solution on a stand-alone basis, the prognostic reports will be rendered through IBM Maximo at a later stage.
Approach: ENRC is applying Cassantec’s prognostic solution for periodical reporting of future malfunction risks and maintenance needs. In a first step, the solution forecasts equipment conditions on the basis of current and historical condition data subject to various operational scenarios. In a second step, future conditions are correlated to equipment-specific malfunction modes, yielding end-of-life forecasts. In a final step, prognostic reports are generated on the basis of end-of-life forecasts, as illustrated in the lower picture. The reports are computed at component level and then aggregated, supporting collective asset management decisions.
Benefits: ENRC is targeting benefits in three main areas:
- Reduction of Downtime Cost: By avoiding lost production from unscheduled delays and by bundling maintenance tasks based on malfunction risk profiles
- Reduction of Maintenance Cost: By extending scheduled maintenance intervals at constant availability and safety levels, and by better preparing for maintenance and replacement tasks
- Transparent Decision Basis: By integrating condition data from different sources, and aggregating insight and foresight at different management levels
Next Steps: The prognostic solution is currently being used at ENRC’s Frontier Mine for the aforementioned types of mining equipment with monthly updates of prognostic reports. The operator is considering an extension of the solution to further equipment, as well as a roll-out of the solution to further mining assets.
– 28 October – written by Moritz von Plate, CEO of Cassantec
The Weather Forecast and Risk Literacy
„There is a 60% chance of rain tomorrow.“
That is the type of information we are getting when looking at our weather app. One might be inclined to shout out: “I don’t want this probability, I want to know whether it will rain or not!”
Despite this apparent lack of relevance, why are we still using weather apps that give us these probabilities? That is, because we have become intuitively risk literate when it comes to weather forecasts.
When I was young, I was told that when the swallows are flying high, the weather will be good. And when they are flying low, it will rain. That heuristic forecast meant that we were soaked during many BBQs, while others never happened despite perfect sunshine.
Nowadays, when we are planning a BBQ, we check our weather app. And it will mention the probability of rain in the coming days. Intuitively, we have become accustomed to incorporate that information into our decisions: At a 60% chance we will probably organize a tent, at an 80% chance we will probably cancel the BBQ and go to the movies instead and at 20% we will probably simply take the risk. If we only want to go for a hike in the nearby park, our decisions will be different, but still be influenced by these probabilities.
In terms of risk analysis, we intuitively consider the following equation:
Risk = Likelihood x Impact.
Without knowing any statistics, I could imagine that users of weather apps make fewer planning mistakes, because they understand this equation. Meanwhile, many operators of industrial assets are still, figuratively speaking, looking at the swallows rather than making use of all the data to generate meaningful risk profiles. They rely on experience and expert assessments rather than on algorithms to compute risk. This means that regularly they get seriously wet, e.g. when a critical gearbox breaks, while their risk-literate colleagues were prepared and have found shelter.
When will these operators will be ready to apply their risk literacy to their professional life?
– October 6 – written by Moritz von Plate, CEO of Cassantec
*** German version below ***
„It’s hard to make predictions, especially about the future!“
What do Niels Bohr, winner of the Nobel Prize in Physics, and Yogi Berra, Baseball Hall of Famer, have in common? The above quote is attributed to both of them.
…and Daniel Kahneman would certainly not disagree as we have seen in my last blog. Now we turn to the 2nd part of the blog: “Expert Intuition: When Can We Trust It?” – further quotes from Daniel Kahneman, „Thinking, Fast and Slow“ (2011), and a short observation by me in the context of the operation of industrial facilities.
- „Intuition as Recognition“: “[…] two basic conditions […]:
• an environment that is sufficiently regular to be predictable
• an opportunity to learn these regularities through prolonged practiceWhen both these conditions are satisfied, intuitions are likely to be skilled.”
- “Indeed, the studies […] never produced a “smoking gun” demonstration, a case in which clinicians completely missed a highly valid cue that the algorithm detected.”
- “If a strong predictive cue exists, human observers will find it, given a decent opportunity to do so. Statistical algorithms greatly outdo humans in noisy environments for two reasons: they are more likely than human judges to detect weakly valid cues and much more likely to maintain a modest level of accuracy by using such cues consistently.”
- “Shortterm anticipation and long-term forecasting are different tasks […].”
- “[…] they have not learned to identify the situations and the tasks in which intuition will betray them. The unrecognized limits of professional skill help explain why experts are often overconfident.”
- The central question is whether the world of industrial assets meets both basic conditions as postulated by Kahneman such as to allow us to generally believe prognoses of the experts. The world of physical assets is certainly not as complex and volatile as the one of financial market experts, whose prognoses are notoriously useless. But our industrial world is probably also less repetitive as the world of fire fighters or nurses, whom Kahneman does grant a certain degree of learnability of prognoses.
- I cannot make a final judgment call to what extent both of Kahneman’s conditions are met.
• Nevertheless, regarding the first condition of sufficient regularity: the increasing complexity along two dimensions (time, content) means that the validity of expert statements meets tight boundaries. An example: the energy transition puts pressure on power plant operators from two directions – while, on the one hand, margins are shrinking and often already negative, the operating demands, on the other hand, are on the rise. The fluctuations in the grid caused by sun and wind power need to be balanced by a load-dependent mode of operation. The impact of that on the life time of plant components is largely unknown such that availability prognoses by experts increasingly resemble a glance into the crystal ball. There are many such examples. They show that the traditional method of expert prognoses delivers insufficient results.
• And regarding the second condition of prolonged practice: in our aging society many companies suffer from a demographically induced loss of experience. This can only partially be mitigated due to cost pressures and the lack of younger successors.
Our take-away is:
The intuition of experts needs the support from algorithms to ensure that Niels Bohr’s statement (or was it Yogi Berra after all?) will finally be remembered as what it really is – a bon mot!
– 06. Oktober – geschrieben von Moritz von Plate, CEO von Cassantec
„It’s hard to make predictions, especially about the future!“
Was haben Niels Bohr, Gewinner des Physik Nobelpreises, und Yogi Berra, Baseball Hall of Famer, gemeinsam? Beiden wird das obige Zitat zugesprochen.
…und Daniel Kahneman würde sicher nicht widersprechen wie wir in meinem letzten Blog gesehen haben. Hiermit kommen wir nun zum 2. Teil des Blogs: “Die Intuition von Experten: Wann können wir ihr vertrauen?” – weitere Auszüge aus Daniel Kahneman, „Schnelles Denken, Langsames Denken“ (2011) und eine kurze Betrachtung von mir im Zusammenhang mit dem Betrieb industrieller Anlagen.
- „Intuition als Wiedererkennen“: „[…]
zwei grundlegende Voraussetzungen [müssen] […] erfüllt sein:
• Eine Umgebung, die hinreichend regelmäßig ist, um vorhersagbar zu sein.
• Eine Gelegenheit, diese Regelmäßigkeiten durch langjährige Übung zu erlernen.Wenn diese beiden Bedingungen erfüllt sind, sind Intuitionen vermutlich sachgerecht.“
- „Tatsächlich haben die Studien […] nie einen Fall nachgewiesen, in dem die Kliniker einen Hinweis mit hoher prognostischer Gültigkeit, den der Algorithmus erfasste, vollständig übersahen.“
- „Wenn ein Hinweisreiz mit hoher Vorhersagekraft existiert, werden ihn menschliche Beobachter finden, falls sie eine geeignete Gelegenheit dafür bekommen. In ‚verrauschten‘ Umgebungen sind statistische Algorithmen Menschen aus zwei Gründen überlegen: Sie spüren mit höherer Wahrscheinlichkeit schwach prädiktive Hinweisreize auf, und sie werden mit viel höherer Wahrscheinlichkeit auf lange Sicht mittelmäßig genaue Vorhersagen liefern, indem sie solche Hinweise konsequent verwerten.“
- „Kurzfristige Antizipation und langfristige Vorhersage sind verschiedene Aufgaben […].“
- „[…] sie haben nicht gelernt, die Situationen und die Aufgaben zu identifizieren, bei denen die Intuition sie im Stich lässt. Die unerkannten Grenzen professioneller Sachkunde erklären, weshalb Experten ihre Fähigkeiten oftmals überschätzen.“
- Die zentrale Frage ist, ob die Welt industrieller Anlagen beide von Kahneman aufgestellten Bedingungen erfüllt, damit Prognosen von Experten generell geglaubt werden kann. Diese Welt ist sicher nicht so komplex und volatil wie die der Finanzmarktexperten, deren Prognosen notorisch unbrauchbar sind. Sie ist aber vermutlich auch weniger wiederholbar als die Welt der Feuerwehrleute oder Krankenpflegern, denen Kahneman eine Erlernbarkeit von Prognosen zugesteht.
- Inwiefern die beiden Bedingungen erfüllt sind, vermag ich nicht abschließend zu beurteilen.
• Dennoch zur ersten Bedingung der hinreichenden Regelmäßigkeit: die zunehmende Komplexität entlang zweier Achsen (zeitlich, inhaltlich) bedeutet, dass der Validität der Expertenaussagen enge Grenzen gesetzt sind. Hier ein Beispiel: Die Energiewende übt auf die Kraftwerksbetreiber Druck aus zwei Richtungen aus – während einerseits die Margen stark rückläufig sind und oft bereits negativ, steigen andererseits die Anforderungen an den Anlagenbetrieb. Die Fluktuationen im Netz, verursacht durch Sonne und Wind, müssen durch Lastfolgebetrieb ausgeglichen werden. Dessen Einfluss auf die Lebensdauer der Anlagenkomponenten ist weitgehend unbekannt, so dass Verfügbarkeitsprognosen durch Experten zunehmend einem Blick in die Kristallkugel gleichen. Derartige Beispiele können beliebig erweitert werden. Sie machen deutlich, dass die herkömmliche Herangehensweise der Expertenprognose unzureichende Ergebnisse liefert.
• Und zur zweiten Bedingung der langjährigen Übung: in unserer älter werdenden Gesellschaft leiden viele Firmen heute schon unter dem demographisch bedingten Wegfall von Erfahrung. Dieser kann aufgrund von Kostendruck und des Mangels an Nachwuchs nur bedingt ausgeglichen werden.
- “The experts performed worse than they would have if they had simply assigned equal probabilities to each of the three potential outcomes. [...] Even in the region they knew best, experts were not significantly better than nonspecialists.”
- “About 60% of the studies have shown significantly better accuracy for the algorithms. The other comparisons scored a draw in accuracy, but a tie is tantamount to a win for the statistical rules, which are normally much less expensive to use than expert judgment. No exception has been convincingly documented.”
- “Several studies have shown that human decision makers are inferior to a prediction formula even when they are given the score suggested by the formula!”
- “They feel that they can overrule the formula because they have additional information about the case, but they are wrong more often than not.”
“Another reason for the inferiority of expert judgment is that humans are incorrigibly inconsistent [...]. When asked to evaluate the same information twice, they frequently give different answers. The extent of the inconsistency is often a matter of real concern.”
A little later: “The brief pleasure of a cool breeze on a hot day may make you slightly more positive and optimistic about whatever you are evaluating at the time.” “Formulas do not suffer from such problems. Given the same input, they always return the same answer.”
- “We knew as a general fact that our predictions were little better than random guesses, but we continued to feel and act as if each of our specific predictions was valid.” (“illusion of validity”)
Cognitive illusions can be stubborn: “when my colleagues and I [...] learned that our […] tests had low validity, we accepted that fact intellectually, but it had no impact on either our feelings or our subsequent actions.”
“Tetlock also found that experts resisted admitting that they had been wrong, and when they were compelled to admit error, they had a large collection of excuses [...]. Experts are just human in the end.”
Reasons for the reaction of the experts:
- “Short-term trends can be forecast [...].”
“[This] contradicts […] everyday experience”: “The problem is that the correct judgments involve short-term predictions […].” “The tasks at which they fail typically require longterm predictions […].”
“They know they are skilled, but they don’t necessarily know the boundaries of their skill.”
“The research suggests a surprising conclusion: to maximize predictive accuracy, final decisions should be left to formulas […].”
- “if a test predicts an important outcome with a validity of .20 or .30, the test should be used.”
- "Die Experten zeigten eine schlechtere Leistung, als wenn sie einfach alle drei möglichen Ergebnisse mit der gleichen Wahrscheinlichkeit eingestuft hätten. [...] Selbst auf dem Gebiet, das sie am besten kannten, waren Experten nicht deutlich besser als Nichtexperten."
- „Bei etwa 60 Prozent der Studien erwiesen sich Algorithmen als erheblich treffgenauer. Die anderen Vergleiche ergaben ein Unentschieden, aber ein Unentschieden ist gleichbedeutend mit einem Sieg für die statistischen Regeln, die im Allgemeinen viel kostengünstiger sind als das Urteil von Experten. Keine Ausnahme wurde glaubhaft dokumentiert.“
- „Etliche Studien haben gezeigt, dass menschliche Entscheider einer Vorhersageformel unterlegen sind, selbst wenn man ihnen das Ergebnis der Formel mitteilt!“
- „Sie glauben, sie könnten sich über die Formel hinwegsetzen, weil sie zusätzliche Informationen über den Fall besitzen, aber damit liegen sie meistens falsch.“
„Eine weitere Ursache für die Unterlegenheit von Expertenurteilen liegt darin, dass Menschen […] in unverbesserlicher Weise inkonsistent sind. Gebeten, dieselben Informationen zweimal zu beurteilen, geben sie oftmals unterschiedliche Antworten. Das Ausmaß der Inkonsistenz gibt häufig Anlass zu begründeter Sorge.“
Weiter hinten dazu: „Die kurze Annehmlichkeit einer kühlen Brise an einem heißen Tag mag dazu führen, dass man zu einer etwas positiveren und optimistischeren Einschätzung gelangt […].“ „Formeln leiden nicht an solchen Problemen. Wenn sie denselben Input erhalten, stoßen sie immer die gleiche Antwort aus.“
- "Wir wussten, dass unsere Vorhersagen kaum besser als ein blindes Raten waren, aber wir fühlten und verhielten uns weiter so, als wäre jede unserer konkreten Vorhersagen gültig." ("Illusion der Gültigkeit")
Kognitive Illusionen können hartnäckig sein: "Als meine Kollegen und ich [...] zur Kenntnis nehmen mussten, dass unsere Tests [...] nur eine geringe prognostische Gültigkeit besaßen, fanden wir uns intellektuell mit der Tatsache ab, aber sie hatte keinen Einfluss auf unsere Gefühle oder unsere anschließenden Handlungen."
"Tetlock fand auch heraus, dass Experten nur widerwillig zugaben, sich geirrt zu haben, und wenn sie gezwungen waren, einen Fehler zuzugeben, hatten sie jede Menge Ausreden parat. [...] Experten sind schließlich auch nur Menschen."
Begründung für die Reaktion der Experten:
- „Kurzfristige Trends lassen sich vorhersagen [...].“
„widerspreche[n] Alltagserfahrungen“: „Das Problem ist, dass die richtigen Urteile kurzfristige Vorhersagen […] umfassen.“ „Die Aufgaben, in denen sie versagen, erfordern typischerweise langfristige Vorhersagen […].“
„Sie wissen, dass sie fachkundig sind, aber sie kennen nicht unbedingt die Grenzen ihrer Fachkunde.“
„Die Forschungsergebnisse legen eine überraschende Schlussfolgerung nahe: Um die Vorhersagegenauigkeit zu maximieren, sollten abschließende Entscheidungen Formeln überlassen werden […].“
- "wenn ein Test ein wichtiges Ergebnis mit einer Zuverlässigkeit von 0,20 oder 0,30 prognostiziert, sollte man ihn anwenden."
- Data input: rather than linking into APIs and retrieving data, Cassantec has adopted a push-approach. Data batches are periodically exported from the customer’s database (typically automatically using small programs, but sometimes also manually) and transferred to Cassantec through several different means (typically upload to a SSL-secured ftp-server, but sometimes also manually through e-mail or on storage devices like CD ROM). The simpler the data format the better it is; .csv or .txt files are best, but all machine-readable formats are accepted.
- Results output: our calculation results are stored in standard JSON files (.txt-format). These can be fed into Cassantec’s proprietary html-based report format or any other format, ranging from legacy binary files or Excel to sophisticated Asset Management software. That way, users accustomed to their preferred ways can benefit from the advanced Prognostic Reports without having to install new and expensive IT (hard- or software).
- It goes without saying that Cassantec’s HTML5/JS-based reports are compatible with current and legacy browser versions.
- Furthermore, the often slow and intermittent access to the Internet at industrial facilities has been taken into consideration in the solution architecture. Results can be cached and used offline and the need for data transfers has been reduced to a minimum.
1) Many decisions are far too complex to be fully digitized in the near future.
2) People can get a great deal of decision support from machines.
3) People are beginning to embrace the change.
1. Build data history: our experience is that most advanced equipment operators gather and archive a plethora of process and condition data; therefore, this step is usually already completed before we come on board.
2. Hand over historical data batch: this is truly low-tech according to today’s standards; we are happy to receive data histories in the csv format on a memory stick, on a CD Rom or uploaded to our ftp site.
3. Gather further information: sufficient are a rough sketch of the equipment (certainly no P&ID or anything with similar detail) and an indication in the sketch where the data sources are located, e.g. where the sensors sit or where the lubricant samples a taken. Not a prerequisite, but a nice-to-have, are your internal warning & alarm levels for the different data sources.
4. Specify malfunctions: for many types of equipment, e.g. many pumps, turbines, fans, compressors, boilers, generators or transformers, we know the typical malfunctions. One half-day workshop with the operator’s engineering, reliability and/or maintenance staff are enough to adjust these to the specific operation at hand. The effort for equipment where our experience is less advanced is not much larger, probably reaching 2 or max 3 half-day workshops.
5. Develop specific data parameters: The same workshops, during which the malfunctions are specified, are used for the definition of the data parameters. Our extensive experience, coupled with the operators’ know-how, allow a finalization of this work-step in the same go together with step (4).
6. Automate data transfer: Although not strictly required for a validation of the prognoses, an automatic data transfer is needed for periodical updates of the Prognostic Reports. This is usually accomplished with a standard data query and – very importantly – it does not require any integration of our software with your historian or other systems.
7. Discuss results, use forecasts: with the completion of step (6), you are ready to verify the prognoses and use the advanced foresight for optimized operations & maintenance decisions.
Unser Fazit daher ist:
Die Intuition von Experten braucht die Unterstützung durch Algorithmen, damit die Aussage von Niels Bohr (oder war es doch Yogi Berra?) endlich nur noch ein Bonmot ist!
– 16. September – co-authored by Moritz von Plate, CEO of Cassantec
***German version below***
Daniel Kahneman: “to maximize predictive accuracy, final decisions should be left to formulas“
Important topics should be left to intelligent persons. That is why I share quotes by Daniel Kahneman, winner of the Nobel Prize in Economics, about his observations on experts vs. algorithms that he made during decades of research.
About the quality of prognoses by experts:
About the reasons for the lack of quality of experts’ prognoses:
About the reaction of the experts:
“Fortunately, the hostility to algorithms will probably soften as their role in everyday life continues to expand.”
This was part 1 on Daniel Kahneman‘s findings.
To be continued with part 2: “Expert Intuition: When Can We Trust It?” – further quotes from Daniel Kahneman, „Thinking, Fast and Slow“ (2011), and a short discussion by me in the context of the operation and maintenance of industrial facilities.
– 16 September – geschrieben von Moritz von Plate, CEO von Cassantec
Daniel Kahneman: „Um die Vorhersagegenauigkeit zu maximieren, sollten abschließende Entscheidungen Formeln überlassen werden“
Wichtige Themen sollte man intelligenten Personen überlassen. Deswegen lasse ich heute ausschließlich den Nobelpreisträger Daniel Kahneman mit seinen Beobachtungen zu Experten vs. Algorithmen, die er im Rahmen seiner jahrzehntelangen Forschungsarbeit gemacht hat, zu Wort kommen:
Zur Qualität von Prognosen durch Experten:
Zu den Gründen für die mangelnde Prognosegüte von Experten:
Zur Reaktion der Experten:
„Glücklicherweise wird die Ablehnung von Algorithmen vermutlich in dem Maße zurückgehen, wie sie im Alltagsleben eine immer größere Rolle spielen werden.“
Dies war Teil 1 zu Daniel Kahnemans Ausführungen.
To be continued mit Teil 2: “Die Intuition von Experten: Wann können wir ihr vertrauen?” – weitere Auszüge aus Daniel Kahneman, „Schnelles Denken, Langsames Denken“ (2011) und eine Diskussion von mir im Zusammenhang mit dem Betrieb industrieller Anlagen.
– 07 September – written by Moritz von Plate, CEO of Cassantec
Internet of Things: the Technical Sophistication of Applications Must Cope with Low Tech Realities
This is Cassantec’s CTO recently near a copper mine in the DR Congo trying to get a signal:
His experience shows that the reality is often-times much more low-tech and driven by legacy systems than IoT evangelists would like it to be. And in order to succeed in this world, solutions need to cope with that.
Through our SaaS approach, Cassantec has managed to couple state-of-the art algorithms and software with tried-and-true IT-features, linking our Prognostic solution seamlessly into the reality of the industrial world.
While the core solution is written in Scala and Java comprising highly advanced stochastic algorithms and data-analysis techniques, its interfaces are cunningly simple and standard:
Through the means outlined above, Cassantec is ready to meet its customers’ needs for state-of-the-art insight, while being simple to implement, it is also flexible and mobile, without having to rely on the latest and greatest the IT-world has to offer.
– 16 July – written by Moritz von Plate, CEO of Cassantec
Human vs. Machine: Is SkyNet Coming?
The world is abuzz with discussions about humans vs. machines. The common wisdom is that the machine will replace humans in many professions. One could say: "digital eats [add any profession you can think of]". That may be true. But I am no prophet; and that is why I won't go out on a limb and make a judgment call as to when and to which extent this will be the case.
However, I do say that for the foreseeable future it shouldn't be "digital eats...", but rather "digital changes…”. It changes the way humans make decisions.
So far, so good…
But what does this mean specifically? Here are three hypotheses as to what is going on:
Hypothesis 1: complex decisions
Conferences about “digital this” and “digital that" all feature at least one speaker who describes situations where algorithms analyze data at lightning speed and automatically make decisions, for which humans neither have the computational power nor speed. That is certainly feasible in cases of high data volumes, yet rather limited complexity in the decision alternatives. A good example are forecasts how many bananas are likely to be sold in which supermarket.
This paradigm hits a wall in more complex situations where cause and effect and the resulting range of options to decide from are less obvious. An increasing vibration of a pump shaft can have many root causes (e.g. the process causes the pump to run dry, and/or the operator tortures the pump with steady operation at 110% of capacity, and/or the lubricant is degrading, and/or it is the wrong lubricant, and/or the shaft has been poorly aligned during last maintenance, and/or anything else?) and, hence, a plethora of alternatives on how to react. My hypothesis is that an automatic decision as to what to do when in order to mitigate the problem is not around the corner.
Picture 1: What works for bananas isn’t necessarily applicable to complex machines like this pump
However, modern algorithms can help humans understand the problem much better, thus preparing them for better decisions. And that brings us to the second hypothesis.
Hypothesis 2: decision support
To this day, many decisions are made based on experience and gut feel. Both are, in a way, implicit analysis algorithms that happen inside people’s brains, often unconsciously. Tools like Prognostics drag much of that implicit analysis out into the light where it becomes objective, comprehensive, transparent, comparable, and repeatable.
No human decision maker can claim all these characteristics. But by using machines to deliver these and combining the machine’s output with the human’s ingenious intuition and capability for complex event processing, the resulting decision can be greatly improved.
Hypothesis 3: embracing change
Yet the challenge is that people need to get used to such a new collaboration with machines. That is no easy task. I remember the days when we would ridicule anybody with a mobile phone. Now I am one of these people who text while walking, simultaneously asking an App for directions to my meeting. It seems that I have changed my attitude a great deal in these 20+ years.
Picture 2: The evolution of the mobile phone
And there are two main reasons for that: (1) it wasn’t a big leap from smoke signs to the latest smartphone, but rather a step-by-step evolution; and (2) it was the insight that change is inevitable, which is why I adopted a “just do it” mentality, learning along the way to truly appreciate the benefits.
When speaking to my customers, I increasingly observe the recognition that embracing the digital paradigm is inevitable as well as the understanding that an evolutionary approach is warranted. Recently, a manager of an industrial corporation said to me: “I am still not quite ready for this. But if we don’t start adopting data analytics tools now, we will be left behind in a few years. So, let’s go for it…”
And that is also my conclusion: Let’s go for it, because SkyNet is far away and the Terminator has only been back in the movies!
- 8 June – written by Moritz von Plate, CEO of Cassantec
Back To The Roots: Predictive vs. Prognostic Or What’s The Difference Anyway?
The terms “predictive” and “prognostic” are synonyms. Merriam-Webster is pretty clear in this regard: to prognosticate is “to foretell from signs or symptoms: predict" (http://www.merriam-webster.com/dictionary/prognosticate).
Bar any nuances, which are lost on me as a non-native speaker, people use these words interchangeably. So why then do we at Cassantec make such a fuss about differentiating these two?
The words’ origins give a clue:
Prediction has Latin origins and is composed of præ-, "before," and dicere, "to say" (http://en.wikipedia.org/wiki/Prediction). In other words it is about fore-telling.
Prognosis, on the other hand, has Greek origins (πρόγνωσις) and means "fore-knowing, foreseeing" (http://en.wikipedia.org/wiki/Prognosis).
We consider fore-knowing to indicate much more certainty than fore-telling.
When transferring this differentiation to our world of the Internet of Things, we emphasize that companies using Cassantec Prognostics will actually know in advance what will happen when; they join the pride of Prognosticators. In contrast, companies relying on Predictive Analytics will obtain a lot less depth about the future. Rather than learning when a malfunction will occur they will merely be given an early warning that a malfunction will eventually hit them – just when remains painfully nebulous.
Therefore, while the differentiation seems to be hairsplitting at first, it does add a relevant degree of semantic differentiation, which we will continue to emphasize.
- 1 June - written by Moritz von Plate, CEO of Cassantec
Demonstrating Cassantec Prognostics – Easy, High-Benefit, Low-Risk
Testing new software solutions is a daunting task. The real value can usually only be assessed after cumbersome, costly and lengthy implementation projects. That creates a hurdle, sometimes an insurmountable one, for organizations to test new approaches.
The recent development of SaaS and cloud offerings has removed a lot of these hurdles. However, for many solutions that closely relate to the physical world, e.g. products from the fields of Predictive Maintenance or Predictive Analytics, a lot of the old hassle remains.
We at Cassantec have gone to great lengths to further reduce the effort of testing us out. The result is a smooth and efficient process to configure our SaaS-solution. These are the steps we take and information we need for our patent-pending configuration methodology:
Last, but not least, and to really make the point, these are the things we do NOT need from you for a configuration of our solution:
• Root causes for each type of malfunction;
• Historical frequency of occurrence of each malfunction;
• Degree of influence of each malfunction on equipment performance;
• Maintenance schedule and work description;
• Information on the historical operating mode of the equipment.
Maybe not readily intuitive, the configuration process in itself generates benefits to our customers:
• A systematic documentation of the knowledge about malfunctions, their relation to data and the interpretation of both (similar to the results delivered by RCM). This is especially valuable in organizations with an ageing workforce where critical know-how is about to retire.
• Gain valuable insight into the quality of the data and whether the data covers the relevant malfunctions or not. This facilitates a much more targeted approach towards tapping new data sources than usual.
• Gain a tangible perspective on the benefits (qualitative and quantitative) to be gained from data analysis.
But most importantly, of course, you are now ready to use Prognostics to reach superior maintenance and life cycle decisions based on hitherto unavailable foresight.
- 4 May - written by the Cassantec Team
Hydropower: Managing Remaining Useful Life (RUL) with Prognostics
The use of complex data analytics in order to control and improve processes is a crucial element of the Internet of Things (IoT). For maintenance and repair activities the use of Big Data analytics is likewise becoming increasingly important. With the help of Cassantec’s data-based Prognostic technology, the future condition of equipment can be determined and RUL can be managed proactively. This creates the foundation for an intelligent maintenance planning & operating strategy.
Creating transparency and objective information about the RUL and the future condition of the plant were the main goals of applying Cassantec Prognostics to a hydroelectric power plant in Switzerland.
The project results went beyond what was expected: Detailed prognoses of the future condition of nine critical components are being updated on a monthly basis. The first prognoses already revealed that a sharp increase in vibration of one particular generator bearing was mainly responsible for the limitation of the RUL of the entire asset. A scenario analysis determined the dependence of the vibration data on the operational regime. It showed that the current operation mode decreased the RUL of the plant below what was necessary to stay within the long-term plan of operations.
Based on this scenario analysis the asset operator decided to adjust the operating mode of the generator to extend the generator’s and therewith the entire power plant’s RUL.
- 1 April - written by the Cassantec Team
Follow The Routines And Thought
Processes in Your Plant Operations
You Are Accustomed to
Next time you are making a decision ask yourself:
- Am I making a forward-looking decision?
- What was my thought process to come to that decision?
- Which tools and information sources did I use when making the decision?
In many cases, the answer to the first question will be “yes”. When making yourself aware of the thought process you just went through, you will identify moments when you were weighing probabilities and expectations around future events. While continuing to apply this thought process, you can consider supporting your implicit probabilities and expectations with explicit hard facts, with Prognostics.
You will probably also have used data and information from various tools and systems. We all know how we get accustomed to using our tools in certain ways and are very reluctant to changing these. Cassantec Prognostics accommodates this by offering to integrate the prognostic information into the front-end of your current tools. This needs to be done once during solution configuration. After that you can continue to use your tools while benefiting from the additional prognostic information.
So, rather than adjusting your approach to decision making, we advise you to adjust your usage of helpful tools to your tried and tested habits, thus allowing you to put your decisions on a sounder footing.
Cassantec Prognostics – it’s easy and intuitive to use!
- 26 February - written by the Cassantec Team
Reliability Engineers are Prognosticators
Any equipment operator knows that much of his thinking revolves around the future.
He routinely makes decisions that refer to the future, or in other words, are prognostic in nature.
For instance, even with a fixed maintenance schedule, he decides…
1. when to schedule the next maintenance intervention, while expecting the equipment to run fine until then;
2. how to scope and prioritize tasks and assets in the next scheduled intervention;
3. whether and how to respond to sketchy condition data that’s appearing on the monitoring screens;
4. how to allocate time and attention to the different assets, based on respective importance and perceived urgency of action.
Hence, in essence, Reliability Engineers, Equipment Engineers and Plant Operators are Prognosticators. While everybody’s goal is to base decisions on sound facts, information about the future is in scarce supply. Consequently, Reliability Engineers, i.e. Prognosticators, have to make do with compromises, because no prognostic information is available to them:
- "Only” information about the current condition of the assets is available;
- Even advanced “predictive” diagnostics is a sophisticated way of assessing the current asset condition;
- They have to rely on fleet statistics to make inferences about individual components;
- They follow OEM recommendations;
- They rely on experienced gut feel.
- Clearly, there is an information gap: no objective, data-based information about the future is available!
There is one company filling that information gap: Cassantec!
The Cassantec Prognostic Reports provide the Prognosticators with the right set of information:
- Prognostic horizon about equipment condition of weeks, months and, in certain cases, years;
- Explicit quantification of the risk of operating equipment at any time in the future;
- Transparent calculation of Remaining Useful Life (RUL) based on individual equipment history;
- Integration of all available data sources (without the need to retrofit the equipment with new sensors);
- Aggregation from component level (e.g. pump, turbine, steam boiler) to plant and fleet level....
This enables the Reliability Engineer to:
- Understand when in future a malfunction is likely to occur;
- Reduce maintenance cost;
- Increase asset availability, especially through avoidance of unplanned downtime;
- Improve maintenance planning;
- Enhance commercial strategies.
Cassantec Prognostics –
helping Reliability Engineers do their job according to the highest professional standards!
- 26 January - written by Moritz von Plate, CEO of Cassantec
The Need to Collaborate
The world is just at the beginning of an upheaval caused by the push of ‘digital’ into the industrial space. While industry players have been deeply involved in IT and software for decades – just look, as an example, at the complex instrumentation & control programming – the recent trend is different and new. Whether we call it the Industrial Internet, the Internet of Things, or Industrie 4.0, it creates a great number of challenges and, even more so, opportunities.
Players as diverse as IT powerhouses, software giants, industrial OEMs, and Telecoms are jockeying for position to build the future backbone of the industrial landscape. We at Cassantec are agnostic to who and which model wins. But we are clearly not indifferent to the fact that Prognostics will be a crucial part of the picture.
At the same time, we know of the breadth of solutions that needs to be developed behind the façade of the current buzzwords. Many diverse building blocks will have to be implemented to deliver the promise of the Digital Factory. Since we will not be venturing too far off our Prognostic path, we are working on getting involved with the providers of the necessary platforms. As a logical consequence we have launched a partner strategy and are currently in talks with a number of highly relevant players regarding cooperation, ranging from supplier-buyer relationships to potentially even a “Cassantec Inside” approach.
These are exciting times and we are truly happy to be part of the game. Stay tuned for more…
- 07 January - written by the Cassantec Team
The benefits of a data-driven maintenance strategy
The list of benefits of using prognostic data analysis for maintenance management is long. Benefits include:
- Knowing when equipment will likely run into trouble
- Proactive prevention of failures is enabled
- The overall system knowledge is increased (the impact of different operating modes on likelihood of malfunctions becomes transparent)
- Unscheduled maintenance can be turned in to well-ahead planned and scheduled maintenance
- Intelligent bundling of maintenance work is enabled, yielding overall less and only technically really warranted interventions
For more information on how to succeed with a data-driven maintenance strategy you may read the recently published blog article of SMRP (Society for Maintenance and Reliability Professionals) about data-driven maintenance strategies.
- 19 December - written by the Cassantec Team
From Reactive and Preventive Maintenance to Prognostics – an Overview
Asset management comprises a large array of approaches, methods, acronyms and the like. With this article we are trying to provide a concise overview of core aspects in the landscape. It is intended to be a living document and will be updated occasionally based on recent developments, comments, and discussions. We are looking forward to your feedback.
Within the broader field of asset management, there are various approaches to maintenance management, which – in many cases – are used side by side depending on failure risk and criticality of the equipment:
Reactive Maintenance (“Run to Failure”)
Reactive maintenance depicts the simple method of fixing a problem after it occurred. The condition of the equipment is not monitored systematically or data-based.
For equipment that is not mission critical and for which downtime is not costly and/or there is enough redundancy to substitute equipment quickly, reactive maintenance may be the way to go. Yet, in operations in which uptime is critical, damages are costly; and where there is low redundancy, reactive maintenance is unsuitable. Also, in case of catastrophic events in case of malfunctions this method is clearly insufficient.
This method is based on a predefined schedule of regular maintenance interventions. The intervals usually follow meantime between failure (MTBF) estimates from similar components or the OEM recommendations.
This method is fairly easy to organize and yields better results in terms of availability than reactive maintenance. It does not, however, prevent malfunctions systematically. Maintenance might be performed too early when there was not yet a technical need for it. Similarly it might have been scheduled too late and failed to prevent a malfunctions. Actually, given that hardly any equipment exactly behaves like the average, there is a high likelihood of maintenance to be done too early or too late.
Condition-based Maintenance (CBM)
CBM depicts the method of performing maintenance when the technical need for it arises. It is based on condition data of the equipment. Maintenance is performed after one or more indicators show that the equipment is going to fail or its performance is deteriorating. Using this information one can perform maintenance at the right time in order to prevent a failure.
Predictive Maintenance (PdM)
For PdM data are used to predict the future condition of equipment in order to indicate when maintenance should be performed. PdM allows conducting maintenance when technically needed. Thus, failures can be prevented and no maintenance is performed when not necessary at this point in time.
The definitions for CBM and PdM are based on their respective Wikipedia articles. Our interpretation is that these definitions essentially describe the same approach. Do you agree with our understanding?
The approaches defined above are sorted by the required level of sophistication regarding data analysis. We are grouping data analysis approaches into the following three usage categories:
Objective and data-based real-time assessment of equipment condition. For this method the condition of the equipment is monitored by data measurement tools. The data are typically stored in a historian. The observed condition is taken into consideration for maintenance decisions.
(Predictive) Diagnostics / (Predictive) Analytics
Objective and data-based analysis of current equipment condition. The term ‘predictive’ is added when there are early warnings before a malfunction is reached, yet without providing an explicit time horizon. Thus, insight into the current condition as well as expectations about the future condition are taken into consideration for maintenance decisions.
Objective and data-based prognosis of future conditions with an explicit time horizon. Specifically, the difference between Prognostics and Predictive Diagnostics / Analytics is that the latter opens a time window towards developing a potential malfunction. Yet it remains unclear, how long it will take until this time window closes again. Will it be in two hours and therewith requires immediate attention? Will it be in two days, or two weeks or rather two months? Prognostics, in turn, does not only open the time window by giving an alert that a malfunction will come up. It gives, in addition, a malfunction risk for any time in the future and therewith defines the risk of continuing operation without a maintenance intervention. The information about expected equipment conditions at any point in time in the future is taken into consideration for maintenance decisions.
The following table shows which approaches to data analysis are required for which approach to maintenance management:
Reactive (“Run to Failure”)
No or very limited data analysis
No or very limited data analysis
No or very limited data analysis
While Reactive maintenance rests on the assumption that no maintenance will be conducted, the resulting need for data analysis is limited to making sure that a failure is detected when it occurs.
Since Preventive maintenance derives its schedule from set rules, it also only makes limited use of data. Of course, given that failure should be avoided, installing a Condition Monitoring system does make sense.
CBM and PdM both require extensive data. Since they entail decisions about what to do when, both diagnostic insight into the current equipment condition and prognostic foresight into the future development of the condition are required. We believe that predictive approaches (see above – no explicit future time window) do not offer sufficient foresight. Therefore, Prognostics (also above – explicit forecast when time window closes) is needed.
During discussions with partners and customers, we hear the term Prognostics used in many different contexts. Here is an attempt at categorizing these:
The underlying methodologies for making any kind of future inference come in a range of facets:
Process-related forecasts are maintenance tools that help scheduling and organizing maintenance processes. For instance work order management systems help to procure the right equipment for certain interventions in time based on information of past interventions. Further, the expected costs and the associated work load can be estimated based on the past information. While the information from these forecasts is highly relevant for asset management, it does not help operators in their journey towards CBM / PdM.
Failure-based forecasts require a history of past failures. Past data trends that are associated with a failure are compared with current data. Using, for example, Cox regression, Weibull fitting or physical modeling allows forecasting future asset conditions. The problem with this approach is that failure data are hardly available given their cost to obtain them. Therefore, these approaches are limited to niche applications.
Condition-based forecasts also base their analyses on past and current data. However, in contrast to failure-based forecasts, no failure history is required in order to prognosticate failures. Instead, stochastic models are used to detect data anomalies. These data anomalies, in turn, are related to potential malfunctions, again using stochastic algorithms or expert assessments. The result is a risk distribution over time (e.g. the risk of malfunction x is 2% tomorrow, 5% in 10 days, …). Given that it is hard to ensure objective, transparent, repeatable, and scalable expert assessments, we prefer stochastic approaches.
Prognostics - ideal for assets without a failure history that ought to remain without a failure history!
- 26 November - written by Julia Heggemann
Prognostics: the future is hidden in our past data
Be it in the US, Germany, Switzerland or East Asia - Prognostics is a novel tool for maintenance management anywhere and requires explanation. We introduce the Cassantec blog to make the concept of Prognostics better understood and write about actual projects to make it more tangible. Of course we will also inform you about the latest product updates and relevant company news.
We keep our first blog entry short and simple by answering the question:
What is Prognostics?
Prognostics provides insight into the future state of assets with an explicit time horizon of typically weeks or months, in special cases even years. While predictive analytics tools give early warnings opening a time window without providing any information on when it will close again, Prognostics focuses on specifying the future moment when the window closes, i.e. when the risk of malfunctions is too high. Thus, Prognostics is a tool created to optimize the maintenance management of industrial assets in order to increase their reliability and availability.
By using Prognostics you will learn when certain assets will likely run into trouble. The prognoses are done for each individual asset: that means it does not rely on averages derived from comparable assets but only on the assets actual data. This way the prognoses are highly accurate and have an excellent prognostic strength.
The Prognostics model uses stochastic methods and sophisticated algorithms. Historical data about the asset´s condition are fed into the model to forecast the availability and the risk of failure. An explicit risk profile shows how the failure risk of an asset develops over time.
The Prognostics model uses stochastic methods and sophisticated algorithms. Historical data about the assets condition are fed into the model to forecast the availability and the risk of failure. An explicit risk profile shows how the failure risk of an asset develops over time.
This way maintenance can be scheduled when really needed, that is not too early, but also before the asset is likely to cause problems. Prognostics also allows to detect malfunctions before they turn into a problem: a scenario analysis reveals the influence of for instance different load scenarios on the development of the malfunction risk. Based on experience even small adjustments in operations can prevent many malfunctions and therewith increase the asset´s Remaining Useful Live (RUL).
Cassantec is the first provider of Prognostics. Being complementary to diagnostics and predictive analytics, Prognostics can be used as a stand-alone data-analysis tool or to make the forecasts delivered by predictive maintenance really actionable.
To sum it up, Prognostics:
- Delivers risk profiles about the future state of an asset
- Uses stochastic methods with sophisticated algorithms
- Is based on historical data of specific assets
- Can be used as a stand-alone tool or complementary to predictive analytics
The goal of this first blog entry was to make the concept of Prognostics clearer and to reduce the ambiguity associated with the word. Details on certain aspects and case studies will follow on a regular basis.
Questions, feedback or comments from your side are appreciated.