Search

This review article systematically summarizes the existing literature on utilizing machine learning (ML) techniques for the control and monitoring of electric machine drives. It is anticipated that with the rapid progress in learning algorithms and specialized embedded hardware platforms, ML-based data-driven approaches will become standard tools for the automated high-performance control and monitoring of electric drives. In addition, this article also provides some outlook toward promoting its widespread application in the industry with a focus on deploying ML algorithms onto embedded system-on-chip field-programmable gate array devices.

Data-driven approaches like reinforcement learning (RL) allow a model-free, self-adaptive controller design that enables a fast and largely automatic controller development process with minimum human effort. While it was already shown in various power electronic applications that the transient control behavior for complex systems can be sufficiently handled by RL, the challenge of non-vanishing steady-state control errors remains, which arises from the usage of control policy approximations and finite training times. This is a crucial problem in power electronic applications which require steady-state control accuracy, e.g., voltage control of grid-forming inverters or accurate current control in motor drives. To overcome this issue, an integral action state augmentation for RL controllers is introduced that mimics an integrating feedback and does not require any expert knowledge, leaving the approach model free. Therefore, the RL controller learns how to suppress steady-state control deviations more effectively. The benefit of the developed method both for reference tracking and disturbance rejection is validated for two voltage source inverter control tasks targeting islanded microgrid as well as traction drive applications. In comparison to a standard RL setup, the suggested extension allows to reduce the steady-state error by up to 52% within the considered validation scenarios.

Improving control dynamics and enabling maximum torque and power conversion for a given electrical drive are important target quantities of drive control algorithms. To utilize the electrical drive to its maximum extent during transient and steady-state operation, a time-optimal continuous-control-set model predictive flux control (CCS-MPFC) for permanent magnet synchronous motors (PMSMs) is proposed. This scheme considers torque and current limits as softened state constraints in the CCS-MPFC's optimization problem to prevent transient overcurrents as well as torque over- and undershoots during time-optimal operating point changes. Furthermore, the overmodulation range including six-step operation can be entered seamlessly to ensure maximum power conversion at high speeds. Fastest transients within the whole modulation range are enabled by a time-optimal harmonic reference generator. Here, the flux reference of the CCS-MPFC is complemented with a harmonic content that enables operation in the overmodulation region. Further, the reference is prerotated during transient operation to attain time-optimal control performance. Extensive simulative as well as experimental investigations for linearly and nonlinearly magnetized PMSMs show that, compared with state-of-the-art methods, time-optimal control performance in the whole modulation range without transient overcurrents as well as torque over- and undershoots can be achieved by the proposed control method.

Predictive maintenance (PdM) has become an important industrial feature. Existing methods mainly focus on remaining useful life (RUL) regression or anomaly detection to achieve PdM in a given application. Those approaches assume monotonic degradation processes leading to a single catastrophic failure at the system's end of lifetime. In contrast, much more complex degradation processes can be found in real-world applications, which are characterized by effects like self-healing or noncatastrophic anomalies. A important example of devices with complex degradation are electromechanical relays. As established PdM solutions failed when applied to a real-world relays degradation data set, the maintenance algorithm for unlabeled data (MAUD) is presented to detect signs of wear and enable a service in time. In detail, MAUD is based on an artificial neural network (ANN), which is trained semisupervised. Experiments with measurement data from 546 relays show that MAUD is superior to various existing methods: The static B10 threshold, which represents the state of the art in relay maintenance, is surpassed by a 17.07 p.p. increase in utilization while reducing failures by 6.42 p.p. Methods based on machine learning, such as RUL estimation and anomaly detection, achieved much lower utilization (up to 31.83 p.p.) compared with MAUD while maintaining the same failure rate.

Multi-motor drive systems (MMDS) combine several drives that work together to fulfill one task. Compared to conventional single-motor drive systems modular product concepts can be realized with MMDS thanks to additional degrees of freedom. Because of their mechanical structure consisting of several shafts, clutches, and gear pairings, complex deflection shapes arise which lead to unintended torque oscillations.To compensate for these torque oscillations a data-driven adaptive multiple-input single-output scheme based on two cascaded recursive least squares estimators is proposed. Here, weights (manipulable amplitudes) for each order (multiple of a fundamental frequency) to be compensated are multiplied with unit amplitude harmonic signals of a reference oscillator and added as reference torques to the torque controllers of the MMDS’s drives. These weights are continually adapted by an online identification of transfer paths and disturbances. Furthermore, the torque contribution of the individual drives concerning the compensation task can be changed at runtime utilizing a weighting matrix as tuning parameter by analytically solving a quadratic program with a linear equality constraint. Hence, the proposed algorithm is suitable for automatic self-commissioning requiring only marginal expert intervention. Experimental investigations prove the compensation capability of the approach whereby a reduction of the output torque’s total harmonic distortion (THD) of up to $80\%$ from $26.8\%$ to $5.4\%$ for a representative operation point is achieved.

This paper proposes a model predictive control scheme with a higher-level open-loop torque control and an underlying continuous-control-set model predictive current control (CCS-MPCC) combined with a space vector modulation and an integrated harmonic reference generator (HRG) for permanent magnet synchronous motors (PMSMs), which is able to utilize the DC-link voltage of a two-level inverter to its maximum during transient and steady-state conditions. By formulating the CCS-MPCC as a quadratic program with the voltage hexagon as inequality constraints, the MPCC's typical dynamic response can be achieved during transient operation. When the overmodulation region is reached, additional current harmonics are induced by the voltage constraints, causing a deterioration of the controller's performance. To overcome this problem a HRG is introduced, which calculates at each sampling instant a reference current that contains these harmonics by solving a boundary value problem, using an overmodulation method under the assumption of steady state.Hence, the HRG performs a model-based current reference trajectory prediction ensuring that the CCS-MPCC can enter the overmodulation region. This allows highest control dynamics even at the voltage constraint while ensuring a seamless transition from linear modulation to overmodulation and six-step operation. The method presented here is able to operate in the entire speed range including standstill. Thus, it is not necessary to switch between different controller frameworks for the constant-torque and constant-power regions. Extensive simulative and experimental investigations for a highly utilized PMSM with significant (cross-)saturation effects prove the viability of the proposed control methodology.

Torque control of electric drives is a challenging task, as high dynamics need to be achieved despite different input and state constraints while also pursuing secondary objectives, e.g., maximizing power efficiency. Whereas most state-of-the-art methods generally necessitate thorough knowledge about the system model, a model-free deep reinforcement learning torque controller is proposed. In particular, the deep Q-learning algorithm is utilized which has been successfully used in different application scenarios with a finite action set in the recent past. This nicely fits the considered system, a permanent magnet synchronous motor supplied by a two-level voltage source inverter, since the latter is a power supply unit with a limited amount of distinct switching states. This contribution investigates the deep Q-learning finite control set framework and its design, including the conception of a reward function that incorporates the demands concerning torque tracking, efficiency maximization and compliance with operation limits. In addition, a comprehensive hyperparameter optimization is presented, which addresses the many degrees of freedom of the deep Q-learning algorithm striving for an optimal controller configuration. Advantages and remaining challenges of the proposed algorithm are disclosed through an extensive validation, which includes a direct comparison with a state-of-the-art model predictive direct torque controller.

Temperature sensing in electric motors is an important task to ensure component protection against excessive heat while maximizing the power and torque capabilities. In order to optimize the dynamic performance limits of a motor during online operation, important motor temperatures must be known in real time. Since temperature measurements are associated with costs and integration efforts, model-based estimation methods became highly relevant. In recent years, many promising contributions have been made to this field leading to a vast literature basis. This paper organizes the literary landscape and gives a bird's-eye overview of the three most important estimation classes: Indirect methods, which track temperature-sensitive electrical motor parameters, and direct methods, namely lumped-parameter thermal networks as well as supervised machine learning. Practical limitations as well as special challenges of the respective methods are emphasized throughout the text. The aim of the paper is therefore to facilitate the entry for interested researchers and engineers in this field as well as to stimulate new research impulses. Hence, open problems as well as future research questions are addressed at the end of this review.

Reinforcement learning (RL) based methods are an upcoming approach for the control of power systems such as electric drives. These data-driven techniques do not need an explicit plant model like most common state-of-the-art approaches. Instead, the control policy is continuously improved solely based on measurement feedback pursuing optimal control performance through learning. While the general feasibility of RL-based drive control algorithms has already been proven in simulation, this work focuses on transferring the methodology to real-world experiments. In the case of electric motor control, a strict real-time requirement, safety constraints, system delays and the limitations of embedded hardware frameworks are hurdles to overcome. Hence, several modifications to the general RL training setup are introduced in order to enable RL in real-world electric drive control problems. In particular, a rapid control prototyping toolchain is introduced allowing fast and flexible testing of arbitrary RL algorithms. This simulation-to-experiment pipeline is considered an important intermediate step towards introducing RL in embedded control for power electronic systems. To highlight the potential of RL-based drive control, extensive experimental investigations addressing the current control of a permanent magnet synchronous motor utilizing a deep deterministic policy gradient algorithm have been conducted. Despite the early state of research in this domain, promising control performance could be achieved.

Micro- and smart grids (MSG) play an important role both for integrating renewable energy sources in electricity grids and for providing power supply in remote areas. Modern MSGs are largely driven by power electronic converters due to their high efficiency and flexibility. Controlling MSGs is a challenging task due to requirements of power availability, safety and voltage quality within a wide range of different MSG topologies resulting in a demand for comprehensive testing of new control concepts during their development phase. This applies, in particular, to data-driven control approaches such as reinforcement learning, of which the stability and operating behavior can hardly be evaluated on an analytical basis. Therefore, the OpenModelica Microgrid Gym (OMG) package, an open-source software toolbox for the simulation and control optimization of MSGs, is proposed. It is capable of modeling and simulating arbitrary MSG topologies and offers a Python-based interface for plug & play controller testing. In particular, the standardized OpenAI Gym interface allows for easy data-driven control optimization. The usage and benefits of OMG for designing and testing data-driven controllers are demonstrated utilizing Bayesian optimization. Both the current and voltage control loops of a voltage source inverter operating in standalone, grid-forming mode for a remote MSG are automatically tuned given an uncertain application environment. Finally, the transfer to real-world laboratory experiments is successfully demonstrated.