Skip to main content

About this book

This book provides a complete picture of several decision support tools for predictive maintenance. These include embedding early anomaly/fault detection, diagnosis and reasoning, remaining useful life prediction (fault prognostics), quality prediction and self-reaction, as well as optimization, control and self-healing techniques. It shows recent applications of these techniques within various types of industrial (production/utilities/equipment/plants/smart devices, etc.) systems addressing several challenges in Industry 4.0 and different tasks dealing with Big Data Streams, Internet of Things, specific infrastructures and tools, high system dynamics and non-stationary environments . Applications discussed include production and manufacturing systems, renewable energy production and management, maritime systems, power plants and turbines, conditioning systems, compressor valves, induction motors, flight simulators, railway infrastructures, mobile robots, cyber security and Internet of Things. The contributors go beyond state of the art by placing a specific focus on dynamic systems, where it is of utmost importance to update system and maintenance models on the fly to maintain their predictive power.

Table of Contents


Prologue: Predictive Maintenance in Dynamic Systems

This introductory chapter intends to provide a general overview about the motivation and significance of predictive maintenance (PdM) in the current literature, its nature and characteristics, as well as the most essential requirements and challenges in PdM systems (Sect. 1). It outlines the main lines of research investigated during the last 20 years in order to cope with the requirements in industrial environments, by identifying and classifying appropriate research directions resulting in methodologies and components already established for and in predictive maintenance systems with a possible smooth transition to preventive maintenance—“what has been done so far” (Sect. 2). Then, it emphasizes on recently emerging challenges that go beyond state-of-the-art, with a specific focus on dealing with dynamic changes in the system and on establishing fully automatized processes and operations (Sect. 3). This serves as a clear motivation for our book, in which most of the chapters are dealing with data-driven modeling, optimization, and control strategies, which possess the ability to be trainable and adaptable on the fly based on changing system behavior and nonstationary environmental influences. The last part of this chapter (in Sect. 3) outlines a compact summary of the content of the book by providing a paragraph about each of the single contributions.
Edwin Lughofer, Moamar Sayed-Mouchaweh

Smart Devices in Production System Maintenance

Smart devices depict a simple and economically efficient way to enhance existing, matured production systems towards a higher degree of digitalization and information exchange. Thus, current studies expect that the market for such devices will undergo a complete change: While today private users purchase most tablets and smartphones, approximately 75% of all mobile devices will be used in the area of industrial automation by 2025. In addition to ramp-up and planning, maintenance of production machines is one of the most promising application areas for smart devices: Lots of information need to be collected, aggregated, and communicated to enable better decisions and more efficient maintenance processes. As more and more sensors and control data become available in networks and clouds, these need to be processed and used. Together with a rising capacity in the area of data analytics, smart devices will thus be a powerful tool for smarter and more competitive maintenance. On the other hand, several obstacles, e.g., regarding system reliability, communication strategy, or ergonomics still have to be overcome to enable a wider application. This chapter therefore introduces the use of smart devices in the area of production systems maintenance.
Eike Permin, Florian Lindner, Kevin Kostyszyn, Dennis Grunert, Karl Lossie, Robert Schmitt, Martin Plutz

On the Relevance of Preprocessing in Predictive Maintenance for Dynamic Systems

The complexity involved in the process of real-time data-driven monitoring dynamic systems for predicted maintenance is usually huge. Up to certain extent, any data-driven approach is sensitive to data preprocessing, understood as any data treatment prior to the application of the monitoring model, being sometimes crucial for the final development of the employed monitoring technique. The aim of this work is to quantify the sensitiveness of data-driven predictive maintenance models in dynamic systems in an exhaustive way.
We consider a couple of predictive maintenance scenarios, each of them defined by some public available data. For each scenario, we consider its properties and apply several techniques for each of the successive preprocessing steps, e.g., data cleaning, missing values treatment, outlier detection, feature selection, or imbalance compensation. The pretreatment configurations, i.e., sequential combinations of techniques from different preprocessing steps, are considered together with different monitoring approaches, in order to determine the relevance of data preprocessing for predictive maintenance in dynamical systems.
Carlos Cernuda

Anomaly Detection and Localization


A Context-Sensitive Framework for Mining Concept Drifting Data Streams

In this chapter, we present the staged learning approach to classification in a non-stationary stream of data. Unlike the standard data stream mining paradigm that assumes change is always present, the staged approach senses the level of volatility in the stream and adjusts the mode of learning accordingly. We propose a scheme whereby volatility could be measured and construct a volatility detector that senses the stream. We model the data stream as consisting of two states: a high-volatility state and a low-volatility state, with transitions taking place to/from these states depending on the level of volatility in the stream. In segments of high volatility an ensemble of online classifiers is used for learning, whereas in low volatility maximum utilization is made of past concepts which are encoded by compact versions of Fourier spectra. The staged approach results in improvements in accuracy as well as throughput while reducing memory usage as demonstrated by our experimentation on a wide range of real-world and synthetic datasets.
Chamari I. Kithulgoda, Russel Pears

Online Time Series Changes Detection Based on Neuro-Fuzzy Approach

The problem of fault detection of time series properties is attractive for many researchers in different areas for a long enough time. The results of the solution of this problem are used in many areas, such as monitoring of the manufacturing processes, control of moving object, bioinformatics, medical diagnostics tasks, and video stream processing. Nowadays, a fairly large number of approaches are proposed for solving this problem. Among popular approaches, there are methods, which are based on statistical analysis of time series, mathematical models of objects that generate these time series, pattern recognition, clustering, and artificial neural networks. The situation is more complicated if the information is fed for processing in online mode, and changes of signal properties can have both abrupt type (faults, outliers, and anomalies) and enough slow drift. At that, these time series can be represented in the vector or matrix sequences form and have not only stochastic character but also chaotic one. In this case, the approach based on computational intelligence methods, first of all, the neuro-fuzzy models with online learning algorithms, can have the most effectiveness. In cases where changes in monitored objects have a smooth slow nature, and as a result, it is impossible to establish a crisp boundary between segments of time series. In this situation, the use of fuzzy clustering methods is effective. At the same time, since algorithms of fuzzy clustering (both probabilistic and possibilistic) are intended to operate in batch mode, their online modifications are proposed, which essentially present the gradient procedures for minimizing conventional fuzzy goal functions. Thus, neuro-fuzzy algorithms are proposed for the fuzzy segmentation of multidimensional time series, which allow detecting in a real time both abrupt and smooth changes in the properties of stochastic and chaotic sequences.
Yevgeniy Bodyanskiy, Artem Dolotov, Dmytro Peleshko, Yuriy Rashkevych, Olena Vynokurova

Early Fault Detection in Reciprocating Compressor Valves by Means of Vibration and pV Diagram Analysis

Leaking valves are the most common reason for unexpected shutdowns of reciprocating compressors. Therefore, a leaking valve has to be detected early and reliable, even for arbitrary operating conditions. In this chapter, a data-driven approach for compressor valve monitoring is proposed. As compressors are equipped with different sensing systems and usually retrofitting new sensors is not desired or even impossible, two independent methods are developed. In the first approach, accelerometers are mounted at the valve covers to perform vibration analysis. In the case of a broken valve, certain time–frequency patterns occur, different from the patterns in the case of varying operating condition. It is thus possible to extract specific features from the time–frequency representation to distinguish between healthy and broken valves. In the second approach, pV diagrams of compression cycles are analysed. Gas flowing through a leak affects the pressure in the compression cylinder and thus the pV diagram. The pV diagram is also affected by varying operating conditions such as load, suction, and discharge pressure. Appropriate features to distinguish these cases are extracted both from the logarithmic pV diagram and the environmental pressure conditions.
Kurt Pichler

A New Hilbert-Huang Transform Technique for Fault Detection in Rolling Element Bearings

Rotary machinery is commonly used in various industries. Most rotating machinery imperfections are related to defects in rolling element bearings. Unfortunately, reliable bearing fault detection still remains a challenging task, especially when bearing defect is at its initial stages, and the defect-related features are nonstationary. A new enhanced Hilbert–Huang transform (HHT) technique, eHT, is proposed in this chapter for incipient bearing fault detection. In the proposed eHT technique, the vibration signal is firstly denoised to reduce impedance effect of the measured vibration signal and enhance signal-to-noise ratio. Then, a morphological filter is proposed using a linearity measure method to demodulate characteristic features from the HHT, and to improve fault detection accuracy. The effectiveness of the proposed eHT technique is verified analytically and experimentally by a series of tests corresponding to different bearing conditions.
Shazali Osman, Wilson Wang

Comparison of Genetic and Incremental Learning Methods for Neural Network-Based Electrical Machine Fault Detection

Learning of neural network structure and parameters using genetic and incremental heuristic algorithms are potential approaches to address the local optima and design issues experienced when using conventional deterministic algorithms and arbitrarily chosen network structures. This chapter presents results on the development of an evolutionary (EANN) and an evolving fuzzy granular (EGNN) neural network for detecting and classifying inter-turns short-circuit in the stator windings of induction motors. A condition monitoring system based on features extracted from voltage and current waveforms associated with a type of neural network is proposed. Both EANN and EGNN are able to develop their parameters and structure according to data samples and a fitness function or error index. Real-programming-based genetic operators, i.e., mutation, recombination, and selection operators, were customized to address the fault detection problem in question. Operators and rates were evaluated in order to obtain a consistent and effective EANN learning algorithm. EGNN handles gradual and abrupt changes typical of nonstationary environment using fuzzy granules and fuzzy aggregation operators. While EGNN is online adaptive, EANN requires a set of initial data. Such initial dataset for training offline neural networks was obtained from a modified induction motor properly designed for insertion of stator shorted-turns and from a fault simulation model useful to extend the dataset by interpolation. Aspects of the neural models, such as classification performance, computational complexity, and compactness, are compared with each other and with the results obtained using a conventional feedforward neural network of similar structure, but trained by a deterministic gradient descent algorithm. The EGNN classifier achieved the best performance on shorted-turn fault detection considering a real dynamic environment subject to voltage unbalance, load variation, and measurement noise.
Daniel Leite

Evolving Fuzzy Model for Fault Detection and Fault Identification of Dynamic Processes

This chapter presents an evolving fuzzy method based on data clouds for fault detection and identification of dynamic processes. The method calculates the local density of the data using the recursive Mahalanobis distance through recursive calculation of the inverse of the covariance matrix. The local density is actually a measure which determines the closeness and the membership degree of the data to the existing data clouds. The structure of the fuzzy model evolves in an online manner and it is capable to incorporate new knowledge in the model. The learning procedure of the model starts with known/labeled data for normal process operation and for faults. Using this knowledge, the method/classifier is capable of identifying the same operation mode the next time it appears. The efficiency of the method is tested on a model of HVAC (Heating, Ventilation, and Air Conditioning) system. The structure of the model was designed directly from a real process, and the model’s parameters were tuned to cope with the dynamics of the real system. Therefore, the model represents the actual behavior of the real system. Besides the overall accuracy of the method, we also tested the efficiency in a manner of true positive and false positive rate. The results were compared to the established statistical fault detection method DPCA (Dynamic Principle Component Analysis).
Goran Andonovski, Sašo Blažič, Igor Škrjanc

An Online RFID Localization in the Manufacturing Shopfloor

Radio Frequency Identification technology has gained popularity for cheap and easy deployment. In the realm of manufacturing shopfloor, it can be used to track the location of manufacturing objects to achieve better efficiency. The underlying challenge of localization lies in the nonstationary characteristics of manufacturing shopfloor which calls for an adaptive life-long learning strategy in order to arrive at accurate localization results. This paper presents an evolving model based on a novel evolving intelligent system, namely evolving Type-2 Quantum Fuzzy Neural Network (eT2QFNN), which features an interval type-2 quantum fuzzy set with uncertain jump positions. The quantum fuzzy set possesses a graded membership degree which enables better identification of overlaps between classes. The eT2QFNN works fully in the evolving mode, where all parameters including the number of rules are automatically adjusted and generated on the fly. The parameter adjustment scenario relies on decoupled extended Kalman filter method. Our numerical study shows that eT2QFNN is able to deliver comparable accuracy compared to state-of-the-art algorithms.
Andri Ashfahani, Mahardhika Pratama, Edwin Lughofer, Qing Cai, Huang Sheng

Prognostics and Forecasting


Physical Model-Based Prognostics and Health Monitoring to Enable Predictive Maintenance

This chapter addresses the development and application of predictive maintenance concepts for several types of assets, following two approaches: (1) detection and prediction of failures based on (real-time) monitoring the health or condition of the systems, and (2) prediction of failures (prognostics) using physical failure models and monitoring of loads or usage. Firstly, several challenges in the field of predictive maintenance are presented. These challenges will be addressed by the methods and tools discussed in the remainder of the chapter. Both the structural health monitoring methods and the prognostic concepts presented are based on a thorough understanding of the system and physical failure behaviour. After discussing the approaches for monitoring and prognostics, a series of decision support tools is presented. As a large number of methods and techniques are available, the selection of the most suitable method, as well as the critical parts in a system, is a challenging task. The presented tools assist in this selection process. Finally, the practical implementation of the presented approaches is discussed by showing a number of case studies in different sectors of industry.
Tiedo Tinga, Richard Loendersloot

On Prognostic Algorithm Design and Fundamental Precision Limits in Long-Term Prediction

System states are related, directly or indirectly, to health condition indicators. This fact has encouraged the development of a series of failure prognostic frameworks based on Bayesian processors (e.g., particle or unscented Kalman filters) to efficiently estimate the Time-of-Failure (ToF) Probability Mass Function (PMF) in nonlinear, non-Gaussian, systems with uncertain future operating profiles. However, the assessment of the effectiveness of these methods has always been a concern for the Prognostics and Health Management (PHM) community. This chapter tackles this issue, providing a formal mathematical definition of the prognostic problem and a rigorous analysis for performance metrics based on the concept of Bayesian Cramér–Rao Lower Bounds (BCRLBs) for the predicted state mean square error (MSE) in prognostic algorithms. Furthermore, a step-by-step design methodology to tune prognostic algorithm hyper-parameters is explored, allowing to guarantee that obtained results do not violate fundamental precision bounds for ToF estimates. The design methodology distinguishes between hyper-parameters that affect the efficiency of the implementation and those that have impact on the efficacy of obtained results, providing a structured procedure to explore different combinations that could improve the characterization of the ToF PMF. It is shown how this design procedure allows detecting situations in which the prognostic algorithm implementation generates results that violate these fundamental precision bounds. In addition, the impact of a relaxation in efficiency constraints on the outcome of the prognostic algorithm is measured, helping the designer to take an informed decision on the hardware that is required to implement the algorithm in real-time applications. These concepts are applied to the problem of End-of-Discharge (EoD) time prognostics in lithium-ion batteries as an illustrative example.
Marcos E. Orchard, David E. Acuña

Performance Degradation Monitoring and Quantification: A Wastewater Treatment Plant Case Study

The performance of electronic and mechanic components used in any industrial process changes over time. The wear generated by uninterrupted usage and the external conditions increase the probability of suffering a failure and also make them less efficient, by reducing their performance and increasing the operational costs. To prevent these consequences, maintenance works are carried out periodically. In order to establish a predictive maintenance plan, it is necessary to have reliable analysis of the performance of these components. For this, a data-driven approach to quantify and monitor the performance degradation of this equipment is presented. This approach is tested over wastewater treatment plant equipment, the blowers from an aeration system and the pumps of two pumping systems. The results obtained with this approach had been validated by the plant managers.
Iñigo Lecuona, Rosa Basagoiti, Gorka Urchegui, Luka Eciolaza, Urko Zurutuza, Peter Craamer

Fuzzy Rule-Based Modeling for Interval-Valued Data: An Application to High and Low Stock Prices Forecasting

Financial interval time series (ITS) describe the evolution of the highest and lowest prices of an asset throughout time. The difference of these prices, the range, is a measure of volatility. Therefore, their accurate forecasts play an important role in many applications such as risk management, derivatives pricing, and portfolio selection, as well as supplement the information by the time series of the closing price values. This chapter proposes an interval fuzzy rule-based model (iFRB) for ITS forecasting. iFRB is a fuzzy rule-based model with affine consequents which provide a nonlinear approach that naturally processes interval-valued data. It is suggested as empirical application the prediction of the main index of the Brazilian stock market, the IBOVESPA. Interval forecasts are compared against traditional univariate and multivariate time series benchmark models and with an interval multilayer perceptron neural network in terms of traditional accuracy metrics, statistical tests, and quality measures for interval-valued data. The results indicate that iFRB method appears as a promising alternative for interval-valued financial time series forecasting.
Leandro Maciel, Rosangela Ballini

Diagnosis, Optimization and Control


Reasoning from First Principles for Self-adaptive and Autonomous Systems

Model-based reasoning or reasoning from first principles is a well-known method for performing various tasks including diagnosis from systems’ models directly. In this chapter, we will first discuss the basic principles and algorithms of model-based reasoning relying on the system models of the correct behavior as well as fault models. Afterwards, we discuss how to provide models including a discussion of the use abstraction. We further extend the basic foundations allowing model-based diagnosis to be applied to self-adaptive systems including fail-operational system. Beside the system architecture comprising monitoring capabilities, we show how to integrate a model-based diagnosis engine enabling the system for reasoning about its internal fault state and for taking appropriate repair or compensating actions after fault localization. We illustrate the underlying concepts using an autonomous mobile robot as example where we focus on the robot’s drive.
Franz Wotawa

Decentralized Modular Approach for Fault Diagnosis of a Class of Hybrid Dynamic Systems: Application to a Multicellular Converter

The majority of real systems are hybrid dynamic systems (HDS) in which the discrete and continuous dynamics cohabit. Their behavior can be described through a set of discrete operation modes and a set of analytical redundancy relations within each discrete mode. The fault diagnosis of HDS is based on the use of a global model. The latter can be too huge to be physically built for large-scale HDS with multiple discrete modes. Therefore, this paper proposes an approach to perform fault diagnosis of HDS, in particular discretely controlled continuous systems, without the use of a global model. In this approach, the system is divided into a set of discrete dynamic components. Then, the local models of the latter are enriched by adding the events generated by the abstraction of the continuous dynamics defined in each discrete mode. For each discrete component, a local diagnoser is constructed based on its enriched local model. Each local diagnoser is sensitive only to faults that impact the behavior (dynamics) of its associated component. Since the local diagnosers are constructed without the use of a global model but only the system discrete components’ local models, this approach scales well to large-scale systems with multiple discrete modes. A three-cell converter is used to illustrate the proposed approach.
Moamar Sayed-Mouchaweh

Automated Process Optimization in Manufacturing Systems Based on Static and Dynamic Prediction Models

A key aspect in predictive maintenance is the early recognition of product downtrends and a proper reaction on it, to reduce production waste and to avoid machine failures, components destruction, and risks for operators. We propose an approach for the automated optimization of process parameters in manufacturing systems in order to automatically compensate possible downtrends in product quality at an early stage. This should significantly reduce or even avoid manual (reaction) efforts for operators which are often time-intensive and laborious. Such downtrends are recognized by prediction models for product quality, which are extracted from process data and which come in two different variants: (1) static predictive mappings established based on process (machining) parameter settings through a combination of a new hybrid variant of design of experiment (DoE), cross-correlation analysis, and data-driven mapping construction; and (2) dynamic forecast models which respect the time-series trends of process values measured during on-line production, being able to properly recognize undesired changes and dynamics happening (unexpectedly) during the process. These models will have the property to be able to self-adapt and evolve over time based on new recordings; they employ generalized (flexible) evolving fuzzy systems (GEFS) combined with a new incremental update of the latent variable space obtained through partial least squares (PLS). Both types of prediction models can then be used as surrogate mappings within a multi-objective, evolutionary optimization process for important target quality criteria, which relies on a fast co-evolution strategy. Several results from a micro-fluidic chip production process will be included to demonstrate the applicability and performance of the proposed methods and to discuss open challenges.
Edwin Lughofer, Alexandru-Ciprian Zavoianu, Mahardhika Pratama, Thomas Radauer

Distributed Chance-Constrained Model Predictive Control for Condition-Based Maintenance Planning for Railway Infrastructures

We develop a Model Predictive Control (MPC) approach for condition-based maintenance planning under uncertainty for railway infrastructure systems composed of multiple components. Piecewise-affine models with uncertain parameters are used to capture both the nonlinearity and uncertainties in the deterioration process. To keep a balance between robustness and optimality, we formulate the MPC optimization problem as a chance-constrained problem, which ensures that the constraints, e.g., bounds on the degradation level, are satisfied with a given probabilistic guarantee. Two distributed algorithms, one based on Dantzig-Wolfe decomposition and the other derived from a constraint-tightening technique, are proposed to improve the scalability of the MPC approach. Computational experiments show that the distributed method based on Dantzig-Wolfe decomposition performs the best in terms of computational time and convergence to global optimality. By comparing the chance-constrained MPC approaches with deterministic approach, and traditional time-based maintenance approach, we show that despite their high computational requirements, chance-constrained MPC approaches are cost-efficient and robust in the presence of uncertainties.
Zhou Su, Ali Jamshidi, Alfredo Núñez, Simone Baldi, Bart De Schutter


Additional information