Fault Diagnosis, Prognosis, and Reliability for Electrical Machines and Drives. Abdenour Soualhi

power electronics, controllers, sensors, and occasionally batteries to operate. This ongoing expansion has resulted from progress on many fronts:

       Electrical machines: new materials, advanced manufacturing processes, and accurate and fast design tools

       Power electronics: new switches with higher energy density, switching characteristics, and efficiency, advanced topologies, continuously improving manufacturing

       Advanced control methods and increasingly powerful computation, both at the operational and at the design level

       Increased collected knowledge from extensive experience

      Mechanical actuators, motion systems, generators, and controls have always relied on Condition Based Maintenance (CBM) to ensure minimal interruptions of service and safety of people, processes, and equipment. Replacing mechanical actuators and controls with electromechanical ones has resulted in higher overall efficiencies, cost savings, and improved performance. It was then natural to improve CBM through the use of health monitoring. It was made possible, with the sensors and signal processing power which are integrated in electrical drives to enable incipient fault identification and diagnosis and to some degree failure prognosis. Toward that end, operating variables have to be monitored, evaluated, and acted upon. To realize these goals, a supervisory system is needed that resides in one or more CPUs. The implementation of this system possibly needs additional sensors installed on some components and systems to filter and process their outputs, a number of internal or external redundant components, subsystems, or systems, a central or distributed controller with algorithms to identify a fault and its severity, predict its development, and take and then implement automatically. What remained an open question and a challenge, at least for early such systems of diagnosis and prognosis, has been whether they result in improvement of reliability. This question and some skepticism, to a large degree, stem from the increased complexity of drive systems, both in the number of parts and the algorithms used, and to a lesser degree from the distrust of the new. Due to the complexities of drives, the number of failure modes has increased, and the ability of the operator to manage them directly has decreased.

The answer to these issues is the development of drive systems that have demonstrably higher reliability while remaining affordable. This means, at best, long, predictable, uninterrupted, as-designed operation, but also the ability to identify any deviation of the performance at present or in the future, and minimally take action to manage it. Besides detection and identification of incipient faults which can result in failure, and diagnosis of the fault severity, it means accurate determination of the expected time before failure, so that appropriate action should be taken. Such action may include continuing normal operation as designed, modification of control, employment of redundancies, or modification of the required performance or mission. To accomplish the tasks associated with diagnosis, prognosis, and reliability enhancement, two components are required beyond the hardware discussed earlier:

       Models of the drive and its components, based on analytical, physics-based models, on collected data from extensive testing of similar components and drives, and even data collected during the drive operation.

       Algorithms based on a large body of diverse work that includes, among others, signal processing, statistical and Bayesian work, and artificial intelligence tools.

      Reliability of a system has been a characteristic that is determined during the design stage, from the reliability of the components, their methods of interaction, and when possible, from past experience of similar systems with the same components. Once decided, the system is launched, and in the classic concept, not modified except through maintenance. Instead, including in the drive at the design stage, a diagnosis and prognosis module integrated with redundancies and decision mechanisms is expected to improve reliability.

      Widespread use of fault diagnosis and failure prognosis and enhanced reliability through these have been elusive goals for many years. Despite extensive research results, applications had been limited to niche markets. It is only recently that related researchwork has translated and resulted in broader adoption in industry. Thousands of scientific papers and many books have preceded this one, discussing in detail the narrow subjects associated with electrical drives, and more broadly, with electromechanical systems. It is the hope of the authors that this book will be a contribution and of use to students and practicing engineers, scientists, and researchers in furthering the application of the rapidly expanding and maturing research results in the interest of improving the safety and the environment.

      The book integrates the results of research efforts, framed by the interests and research of the authors, and consists of a discussion of basic tools, applications and to specifics: power electronics, capacitors, batteries, sensors, electrical machines, and closes with the integration of fault diagnosis and failure prognosis to the enhancement of the reliability of electrical drives.

The first chapter deals with the components involved in the diagnosis and prognosis of breakdowns. It presents the different sensors, the signals obtained, their conditioning, and the different methods using the information extracted from these signals. Thus, time, frequency domain, or time frequency representation could be used to build relevant information. Classifiers, neural networks, rain flow, etc., estimate the present or future state of the system from this information.

      The second chapter deals with the different components of the electrical drives from the supply to the electrical actuator: static switches, capacitors, asynchronous, and synchronous motors. A presentation of their physics and their different faults is introduced before detailing some applications in the diagnosis and prognosis domain.

The last chapter addresses the use of fault diagnosis and failure prognosis discussed in the previous chapters in the improvement of the reliability of electrical drives. Going beyond the classical definitions of reliability and its estimation at the design stage, it presents how failure prognosis and decision systems enhance reliability. It proposes additional measures of reliability based on the cost of fault management and of failure prognosis, as well as of the necessary modification of the mission profile.