Energy, Process & Utilities

Improving Predictive Maintenance by
Leveraging Engineering Technologies

November 2013
Philippe AUDRAIN
Energy, Process & Utilities
Business Consultant Director
Dassault Systemes

Predictive maintenance is the way to apply this strategy toward equipment health. production assets are operated on a 24x7 basis all year long. including equipment health. This paper explores new ways to approach the end-to-end challenge by introducing new techniques from the engineering world. It also highlights the fact that none of these techniques is relevant in every situation. This paper summarizes the various techniques that have been successfully tested and applied in the past decades. Plant Asset Information Lifecycle Management 2 . For several years operators have put maintenance strategies in place to optimize uptime and minimize any type of shutdown. A plant business unit might stay profitable or not just because of the availability of equipment.Executive Summary In today’s highly competitive global economy plant operators in the Energy. The focus is to put under control all parameters of a production process. and by combining several other technologies. Process & Utilities industry are under constant pressure to maximize output while minimizing costs. Operational costs represent almost 60% of overall asset cost. In this industry. but cannot provide efficient solutions in all possible scenarios. each technique can be partially applied. The ultimate target is to be in a position to predict any future behavior as soon as one is perfectly aware of the current situation.

with additional improvement in technology (both IT and captors). But still the maintenance strategy was not optimized. It tries to create a model of normal behavior and then derives the abnormal tendencies. The took advantage of information gathered from captors to set up alerts that trigger inspection and/or work orders to prevent a failure before it happened. Several improvements were then made to achieve a real predictive approach taking into account the available information gathered in-situ. Rather than immediately going on-site to the equipment when receiving an alert. Dedicated solutions were developed to handle this activity as well as possible and with the lowest cost impact. new strategies appeared to lower spare part turn around. They initially focused on managing spare parts and scheduling work orders using simple algorithms aligned with a curative approach and some preliminary preventive measures. Condition-based inspection and conditionbased maintenance appeared. more advanced approaches were used to monitor long-term trends on parameters and that could anticipate the next future failure.Current Predictive maintenance strategies do not fulfill overall Plant owner needs Equipment investment has a huge impact on the cost of final product in all heavy CAPEX industries such as Energy. it still had a big impact on final production cost. Consequently. the discipline of plant and equipment maintenance has played a very important role in the last 30 to 40 years within plant organizations. cyclic inspection was introduced to try and reduce the turn around of spare parts. This first step significantly improved the quality of maintenance by reducing the curative ratio. independent of the number of samples. • This approach excludes the production schedule (so prediction of failure time and related replacement request might impact production). but it has some constraints and inefficiencies: • The ability to collect relevant information at a relevant frequency to get efficient results. However. • This is not a “lifelike approach”—it looks for trends and correlations among parameters. The second stage in the late 1980s combined CMMS with MRP/ERP (Manufacturing Resource Planning and Enterprise Resource Planning) for managing the production schedule to minimize the negative impact of equipment maintenance. The “lifelike approach” is to look at directly abnormal cases/tendencies. Rather than a pre-defined cyclic part replacement (pure preventive approach). Then in the 1990s. • The ability to analyze on-time the collected information to do efficient prediction. CMMS (Computerized Maintenance Management Systems) were introduced in the late 1970s to manage maintenance activity in parallel with operations. This is still the most current approach. Process & Utilities segments. Plant Asset Information Lifecycle Management 3 .

but also the controllers..” • To test enough situations to feel confident of coverage and scope and/or that the model is accurate enough to be used in “extrapolation mode” on other untested situations. etc. it should be possible to integrate the 3D aspect of those assets and its impact on their behavior. yield) • Plant maintenance (to minimize failure) Various approaches to build a reliable behavior model (process or equipment) 1. “White box” explicit mathematical approach: •  Physical equations model-based. The “implicit” approach requires more complex algorithms.. Empiric/heuristic/symbolic history analysis (the most common on-going option to do predictive work): •  Pure informal knowledge capture (“experience”) •  Statistical analysis 2. Alternatively (if no model is available) a pure stochastic black box approach can be used on nondeterministic phenomena which leverages past experiences to identify some behaviors that are potentially reproducible. but is only valid on “short future situation” (shorter “dt”). The “explicit” approach is more precise. if available •  Rules-based (expert system. • To validate and benchmark using well known/documented previous real situations to check the accuracy of prediction on those “reference cases. but can potentially be applicable for longer “dt” prediction.The Basics of Prediction and the various predictive technologies that apply A plant is made up of a set of interconnected equipment/systems as well as attached “controllers/control systems”. pattern recognition) •  Decision tree/Fault tree (for diagnosis support) 3. The incredible increase of IT and automation capabilities in the last 20 years allowed the development of predictive strategies to enhance either: • Plant operation (to optimize throughput. “Black box” explicit mathematical approach (mainly to describe a chemical/transformation process for advanced real-time control strategies): •  “Neural networks” •  SVM (Support Vector Machine) •  Multivariable process step (and/or impulse) response model (mainly used for advanced control of continuous process) 4. Efficient tools should be able not only to model plant equipment by themselves. heat. And ultimately. 3D multiphysics dynamic simulation • Takes into consideration the 3D nature of equipment and how they react to their 3D physical environment (3D influence is not included in the previously listed approaches). The “predictive approach” is natural human behavior. Some definitions and principles Prediction means being in a position (at a time = t) to describe the future status of a system and equipment (at a time = t + dt). • Focuses on equipment behavior versus physical constraints (vibration. This implies to have an explicit or implicit model of the system and equipment (to calculate or extrapolate the future state with a reasonable level of confidence) requires the following: • To build the model and knowledge.) Plant Asset Information Lifecycle Management 4 .

With DELMIA OI (Operation Intelligence) technology. By setting up alerts in time. With EXALEAD SBA (Search Based Application) environment. The analysis of this mass of data allows companies to identify trends and extrapolate the future state of an on-going process. • Prescriptive analytics combines big data. Plant operators are leveraging the huge quantity of “real-time” and historical data captured by captors. This technology can extract “hidden rules” from a series of values related to various parameters that have a priori no strong interactions together. or predict a future event. Both domains (engineering on one side and operation and maintenance on the other side) require reliable models to be able to simulate a future process. DCS (Distributed Control Systems) and SCADA (Supervisory Control And Data Acquisition) or Historians. prescriptive analytics capabilities can be provided in a maintenance scenario. and machine learning to empirically reveal the origins of the most complex problems. filter. mathematical sciences. but can also be (and already are) successfully used for operation and maintenance purposes. and extract meaningful information from massive collections of data with extended dashboarding capabilities. and then suggests decision options to solve them. logic. replay a past event. it is easy and very efficient to sort. this is the most commonly used approach today to do predictive maintenance. Empiric/heuristic/symbolic/stochastic approach As already stated. • Rule learning engine based on a supervised stochastic learning algorithm • “what-if” simulation scenario provides guidelines to reduce risks Plant Asset Information Lifecycle Management 5 . a past breakdown can be avoided.Dassault Systèmes brings multiple technologies that apply to various prediction modeling approaches Dassault Systèmes has a huge portfolio of technologies that not only apply to engineering activities.

However. and shutdown.White Box mathematical approach Dassault Systèmes provides multiple modeling technologies. SD9/FT9 supporting dysfunctional model approach SD9 (Safety Designer) is another modeling technology. Modelica can be used not only during preliminary studies of new plants to size equipment. Isight can also be used to automate SIMULIA (or any other simulation tool) tests. Plant Asset Information Lifecycle Management 6 . we take care of geometry using FEM (Finite Element Method). It also enhances the productivity of such activity by applying various optimization methods (such as “plan of experience”) to achieve the best results. Dymola supporting functional model approach Dymola is based on the Modelica open source language. heat exchangers. By using SD9 you can determine the final consequence of a failure within a system. as well as to diagnose material degradations. the 3D geometry may influence the behavior of those assets in their environment. You can combine SD9 with FT9 (Fault Tree analyzer). based on another language called AltaRica. such as boilers. It provides a functional system with modeling capabilities applicable to all types of complex equipment systems (or a full plant which is a “system of systems”). SIMULIA brings multiphysics dynamic models to simulate the behavior of a piece of equipment facing physics constraints such as vibration or heat. as well as on the “modified geometry” of old equipment (including the 3D defects related to the previous life of the equipment) to predict its remaining lifetime before a dramatic failure. SIMULIA Isight: automation of tests and activities This workflow engine is able to automate a series of tests and simulations executed with multiple tools. SD9 can predict the propagation of the failure within the interconnected components and deliver a final status. etc. It allows you to build system models with dysfunctional description (using failure modes such as FMEA and MTBR). Modelica is widely used in the power generation industry and has many available libraries of components and standard equipment. but also during the actual life of plants to study dynamic behavior of the plant itself during start-up. transient phases. a complementary solution that enhances the efficiency of the overall approach. compressors. You can apply those constraints on the geometry of new equipment. startup and shutdown •  Find optimized operating points and optimization method for operators •  Integrate the model in operating or engineering software (local and remote monitoring) •  Control system validation 3D multiphysics dynamic simulation approach SIMULIA Abaqus supports a 3D multi-physics dynamic model approach All the previous approaches are “1D” methods—they do not take into account the 3D characteristics of equipment. Example: Purpose of doing a Dynamic Modeling a combine cycle power plant •  Check precisely the performances and the design given by manufacturers (commissioning) •  Verify and validate by simulation the scenario of a GT trip. With SIMULIA Abaqus.

Combining several approaches allows companies to cover a wider spectrum of events. etc. starting with failure modes. such as the Eurosyslib project initiated by the European consortium of companies in the nuclear sector. it is possible to combine: • A top-down physical model-based damage prediction approach • A bottom-up semantic network (fault tree) diagnostic support approach Instruction for maintenance on a simulated operating plant Figure 1 One dimension functional and dysfunctional system analysis This increases efficiency and allows companies to perform efficient predictive safety analysis. using a with prescriptive or predictive approach.). Plant Asset Information Lifecycle Management 7 .Why and how to combine several technologies to improve predictive maintenance Each approach and technology has its own area of optimal efficiency regarding failure prediction (for example. which is critical when dealing with events that are not known in advance. For instance.

3DSWYM. Dassault Systèmes’ collaborative solutions foster social innovation. For more which is based on a single 3DEXPERIENCE platform and merges various applications with these benefits: • Directly enabling predictive maintenance (such as the ones listed above) • Supporting adjacent processes (such as virtual training of maintenance operators. GEOVIA. SOLIDWORKS. and provides a rich portfolio of industry solution experiences. Shinagawa-ku. NETVIBES. visit the Dassault Systèmes website at www.3ds. serving 12 industries. Its world-leading solutions transform the way products are designed. The group brings value to over 170. expanding possibilities for the virtual world to improve the real world. EXALEAD. 3DVIA are registered trademarks of Dassault Systèmes or its subsidiaries in the US and/or other countries. visit www. ENOVIA. rue Marcel Dassault CS 40501 78946 Vélizy-Villacoublay Cedex France Asia-Pacific Dassault Systemes K. all rights reserved.000 customers of all sizes in all industries in more than 140 countries. Americas Dassault Systèmes 175 Wyman Street Waltham. Dassault Systèmes.K. and supported.3ds. the 3DExperience Company. Figure 2 Elements of the Safe Plant Operation solution experience Dassault Systèmes recommends the Safe Plant Operation industry solution experience. spare part intelligence) This environment will change the game of predictive maintenance. produced.© Dassault Systèmes 2014. ThinkPark Tower 2-1-1 Osaki. Our 3DEXPERIENCE® platform powers our brand applications. provides business and people with virtual universes to imagine sustainable innovations. plant asset information lifecycle. Massachusetts 02451-1223 USA Europe/Middle East/Africa Dassault Systèmes 10. SIMULIA. Tokyo 141-6020 JAPAN . For details about the Safe Plant Operation industry solution experience. CATIA.