Search

Most traction drive applications lack accurate temperature monitoring capabilities, ensuring safe operation through expensive oversized motor designs. Classic thermal modeling requires expertise in model parameter choice, which is affected by motor geometry, cooling dynamics, and hot spot definition. Moreover, their major advantage over data-driven approaches, which is physical interpretability, tends to deteriorate as soon as their degrees of freedom are curtailed in order to meet the real-time requirement. In this article, deep recurrent and convolutional neural networks (NNs) with residual connections are empirically evaluated for their feasibility on predicting latent high-dynamic temperatures continuously inside permanent magnet synchronous motors. Here, the temperature profile in the stator teeth, winding, and yoke as well as the rotor's permanent magnets are estimated while their ground truth is available as test bench data. With an automated hyperparameter search through Bayesian optimization and a manual merge of target estimators into a multihead architecture, lean models are presented that exhibit a strong estimation performance at minimal model sizes. It has been found that the mean squared error and maximum absolute deviation performances of both, deep recurrent and convolutional NNs with residual connections, meet those of classic thermodynamics-based approaches, without requiring domain expertise nor specific drive train specifications for their topological design. Finally, learning curves for varying training set sizes and interpretations of model estimates through expected gradients are presented.

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.

This article presents an approach to the controller design for electrical drives, which makes use of methods of deep reinforcement learning. Conventional control methods dominated the field for a long time, since they usually lead to control solutions with very robust and steady results. Yet, it often can be found that the overall control performance heavily correlates with the experience and education of the developing engineer. Moreover, conventional methods strongly depend on the available knowledge of the control system (e.g., plant model accuracy), which often causes the necessity for thorough identification methods. Real-time capability issues are also a present problem of sophisticated control approaches, such as model-predictive methods. Especially, in the domain of electrical drive train control, solving elaborate online optimization problems may be critical when very small plant time constants have to be considered. The methods of deep reinforcement learning will not only enable to acquire a suitable controller structure, but, moreover, the procedure will tune itself, which will allow for a more abstract level of investigation. This article presents a first proof of concept by means of controlling the phase currents of a permanent magnet synchronous motor in a field-oriented framework. The results found are promising and motivate further research in this field.

In this paper, a new temperature observer topology is presented which overcomes the shortcomings of previous ones and achieves a higher accuracy, and a more robust disturbance rejection. It makes use of the Gopinath-style flux observer and combines a lumped-parameter thermal network operating at low speeds and a flux-based permanent magnet temperature observer operating at medium and high speeds. Simulation and experimental results on a 50 kW permanent magnet motor show a performance enhancement over standard topologies; particularly, a superior disturbance rejection to voltage estimation errors. A detailed analysis of the optimal controller tuning is also presented. Furthermore, a Kalman filter is incorporated to account for sensor noise and model uncertainties. Experimental results show an effective fusion of independent temperature estimation methods leading to a superior accuracy compared to the previously investigated approaches. Moreover, the Kalman filter-based fusion offers the capability of detecting temperature-related system failures, e.g., cooling circuit malfunctions.

Accurate temperature estimation of parts of electric machines using a lumped-parameter thermal network (LPTN) requires knowledge of power loss and thermal parameters. However, modeling parameter varying model inputs (power loss for each node and convective thermal conductances) based on physical equations using only measurement signals available during normal drive operation as inputs is not tangible. Hence, a black-box feed-forward neural network (FFNN) based model is introduced as part of the low-order LPTN which is parametrized using empirical measurements dataset using a newly introduced step-by-step system identification instead of a global identification. Then, the proposed hybrid model performance is evaluated using three different unseen cross-validation data sets, which granted an average maximum absolute error of 5 K.

Due to the extensive use of induction motors in torque-controlled applications, e.g. in electric vehicles, high torque estimation accuracy is of paramount research importance. Traditionally, the performance of torque estimation highly depends on the accuracy of the observed magnetic flux which is a challenging task due to various non-ideal effects like magnetic saturation, iron losses or skin effect influences. In contrast, a hybrid machine learning observer for torque estimation is presented which also indirectly serves as a stator flux observer, i.e., flux and torque estimation are combined in one approach. To achieve both high estimation accuracy and small model size, not arbitrary neural network topologies are used, but physically-inspired structures based on expert knowledge (hybrid modeling). The main advantage of this method is that the training of the contained neural networks relies solely on recorded torque measurements and no additional flux measurements are required. In the complete operating range, this approach leads to a root mean square torque estimation error of only 1.0 % related to nominal torque. For comparison, observing the magnetic flux with a standard open-loop current model results in a normalized root mean square torque error of 4.6 %.

Both high-precision and high-efficient torque control of induction motor drives is an important research field due to the extensive use of these motors in torque-controlled applications, e.g. electric vehicles. To achieve high precision, a gray-box rotor flux observer and inverter model have been lately introduced and validated to be effective. Gray-box models (GBMs) combine first-order principles from physics with data-driven identification enabling accurate model performances at low model complexity. Since the effectiveness of an operating strategy significantly depends on the accuracy of the underlying loss model, in this paper, the aforementioned GBMs’ scope is extended to estimate also the power losses in the motor and inverter. Hence, the achieved universal drive model delivers flux, torque, and loss estimations which are substantial for various control tasks, like torque control, operating strategy, or thermal models. In comprehensive test bench investigations, which take into account the entire drive operating range, the measured power losses are compared to the model estimations. This analysis validates that the GBMs can estimate the losses in the motor with a root-mean-square error of 0.47 % and in the inverter with 0.87 %, both related to the nominal mechanical power of the motor.

Electric motors are used in many applications, and their efficiency is strongly dependent on their control. Among others, linear feedback approaches or model predictive control methods are well known in the scientific literature and industrial practice. A novel approach is to use reinforcement learning (RL) to have an agent learn electric drive control from scratch merely by interacting with a suitable control environment. RL achieved remarkable results with superhuman performance in many games (e.g., Atari classics or Go) and also becomes more popular in control tasks, such as cart-pole or swinging pendulum benchmarks. In this work, the open-source Python package gym-electric-motor (GEM) is developed for ease of training of RL-agents for electric motor control. Furthermore, this package can be used to compare the trained agents with other state-of-the-art control approaches. It is based on the OpenAI Gym framework that provides a widely used interface for the evaluation of RL-agents. The package covers different dc and three-phase motor variants, as well as different power electronic converters and mechanical load models. Due to the modular setup of the proposed toolbox, additional motor, load, and power electronic devices can be easily extended in the future. Furthermore, different secondary effects, such as converter interlocking time or noise, are considered. An intelligent controller example based on the deep deterministic policy gradient algorithm that controls a series dc motor is presented and compared to a cascaded proportional-integral controller as a baseline for future research. Here, safety requirements are particularly highlighted as an important constraint for data-driven control algorithms applied to electric energy systems. Fellow researchers are encouraged to use the GEM framework in their RL investigations or contribute to the functional scope (e.g., further motor types) of the package.

The term microgrid is used to describe the concept of a local grid consisting of energy sources, storage facilities and consumers in various energy sectors, which operates with or without external grid connection. This structure creates a variety of flexibility options in operation. The overwhelming majority of studies in the literature work with abstract simulation models on a conceptual level. However, the transfer of the very promising results into real applications and industrial products is still associated with great uncertainties and risks. To fill this gap, the Microgrid Laboratory (μG-Lab) is being established which will help validating microgrid control and operation schemes in a flexible, real but isolated system. After a brief overview of microgrids the components of the μG-Lab will be described. The μG-Lab is a research infrastructure that is being installed until 2022.

High precise torque estimation and torque control for induction motor drives is still an important research field due to the extended use of induction motors in torque controlled applications, for example in electric vehicles. Since the performance of torque estimation highly depends on the accuracy of the observed magnetic flux, a Kalman filter for rotor flux observation in combination with a global parameter identification is presented in this paper. Thereby, the basic induction motor model is extended by magnetic saturation of the main inductance, by an iron loss resistance, and by skin effect influences within the rotor and stator resistance. For estimating the set voltages of the inverter, an accurate gray-box inverter model is utilized. The parameters of the motor model are globally identified by a particle swarm optimization (PSO). For training and validation a comprehensive data set from test bench measurements is utilized. With this approach, in the complete operation range, a root mean squared torque estimation error of only approx. 0.8 % with respect to the nominal torque can be achieved. For comparison, a simple open- loop current model observer is additionally presented, which takes into account magnetic saturation and skin effect influences, and whose parameters are identified by PSO. This observer leads to a normalized root mean squared torque error of 2.9 %.

Permanent magnet synchronous motors (PMSMs) are traditionally operated in the field-oriented control (FOC) framework based on rotor-fixed coordinates. To set up a proper torque control in this framework, numerous motor parameters are required (e.g. inductance matrix) and, therefore, engineering effort is needed to extract the parameter values from experimental testing or finite element analysis. Moreover, the torque plant model becomes coupled and nonlinear. While the effort to handle these issues is usually acceptable in premium drives (e.g. in electrical vehicles), low-cost applications require control approaches with minimum commissioning complexity. To achieve that objective, this contribution proposes a very simple stator flux-oriented FOC framework requiring only limited parameter knowledge. As a positive side-effect, the proposed framework is operating inherently position-sensorlessly. Experimental tests in the constant torque and flux weakening area proof the principle functionality of the proposed method, but also limitations are highlighted (e.g. decreased torque accuracy). The found results motivate further investigations in this field.

Permanent magnet synchronous machines (PMSMs) are a popular choice in many traction drive applications due to their high energy and power density and moderate assembly costs. However, electric motor thermal robustness in general is harmed by the lack of accurate temperature monitoring capabilities such that safe operation is ensured through oversized materials at the cost of its effective utilization. Classic thermal modeling is conducted through lumped-parameter thermal networks (LPTNs), which help to estimate internal component temperatures rather precisely but also require expertise in choosing model parameters and lack physical interpretability as soon as their degrees of freedom are curtailed in order to meet the real-time requirement. In this work, it is shown that, as an alternative to LPTNs, linear regression achieves similar predictive performance with low computational complexity as long as input representations are preprocessed with exponentially weighted moving averages. Thus, domain knowledge becomes neglectable, and estimation performance depends entirely on collected data and considered input representations. Furthermore, dependence on data quantity and data diversification is examined in order to assess the minimal mandatory amount of test bench measurements.

Speed-controlled induction machine (IM)drives are the working horses in many industrial applications. When applying model predictive control (MPC)to such tasks, two major challenges have to be faced: First, an accurate and dynamic load torque estimation is required for sufficient steady-state and transient control performance. Second, the speed and the underlying current control loop show widely differing time-constants and, therefore, designing a control structure with sufficient prediction horizons for both control variables while ensuring real-time capability on typical hardware platforms is crucial. To overcome the latter problem, a hierarchical MPC approach with different sampling intervals for speed and current control is proposed. Furthermore, a moving-horizon estimator is seamlessly introduced within the speed MPC structure for suitable load torque computation. The proposed control framework is implemented on a rapid-control prototyping test bench utilizing the ACADO toolkit to solve the underlying optimization problems. The method's feasibility is demonstrated by experimental investigations on a laboratory test bench.

Most traction drive applications using permanent magnet synchronous motors (PMSMs) lack accurate temperature monitoring capabilities so that safe operation is ensured through expensive, oversized materials at the cost of its effective utilization. Classic thermal modeling is conducted with e.g. lumped-parameter thermal networks (LPTNs), which help to estimate internal component temperatures rather precisely but also require expertise in choosing model parameters and lack physical interpretability as soon as their degrees of freedom are curtailed in order to meet the real-time requirement. In this work, deep recurrent and convolutional neural networks with residual connections are empirically evaluated for their feasibility on the sequence learning task of predicting latent high-dynamic temperatures inside PMSMs, which, to the authors' best knowledge, has not been elaborated in previous literature. In a highly utilized PMSM for electric vehicle applications, the temperature profile in the stator teeth, winding, and yoke as well as the rotor's permanent magnets are modeled while their ground truth is available as test bench data. A model hyperparameter search is conducted sequentially via Bayesian optimization across different random number generator seeds in order to evaluate model training consistency and to find promising topologies as well as optimization strategies systematically. It has been found that the mean squared error and maximum absolute deviation performances of both, deep recurrent and convolutional neural networks with residual connections, meet those of LPTNs, without requiring domain expertise for their design. Code is available at [1] to assist related work.

A key aspect in the control of electrical drives is the demand for a high control performance. In addition to a short settling time and a small setpoint deviation, a low current ripple is desired. The unavoidable ripple is caused by the switching principle of the used voltage-source inverters. The consideration of the ripple when determining the actuating variables differs between the established control approaches such as the field-oriented control (FOC), the finite-control-set (FCS), or the continuous-control-set (CCS)model predictive control (MPC). This also depends on whether a dedicated modulator is used or not. A CCS-MPC with integrated modulator which takes the principle of a pulse width modulation-like modulator (PWM)during optimization into account is presented. The setpoint deviation and the current ripple are the objectives of a optimization problem. Here, choosing an appropriate cost function is investigated such that the performance in steady state is comparable to that of a FOC and during transients to that of a MPC approach. The chosen cost function also defines the class of the optimization problem to be solved online in each controller cycle. Besides a simulative investigation, an experimental setup with a permanent magnet synchronous motor (PMSM)and qpOASES as solver for real-time optimizations was implemented.

Monitoring critical temperatures in permanent magnet synchronous motors (PMSM) is essential to prevent device failures or excessive motor life-time reduction due to thermal stress. A lumped-parameter thermal network (LPTN) consisting of four nodes is designed to model the most important motor parts, i.e., the stator yoke, stator winding, stator teeth, and the permanent magnets. An empirical approach based on the comprehensive experimental training data and a particle swarm optimization are used to identify the LPTN parameters of a $\mbox{60}$ -kW automotive traction PMSM. Varying parameters and physically motivated constraints are taken into account to extend the model scope beyond the training data domain. Here, a so-called global identification technique for linear parameter-varying systems is innovatively applied to a thermal motor model for the first time. The model accuracy is cross-validated with independent load profiles, and a maximum estimation error (worst-case) of $\mbox{8}\,^\circ$ C regarding all considered motor temperatures is achieved. Also, a comprehensive residual statistical analysis proves suitable estimation results in terms of model robustness and accuracy.

The control performance of three-phase electrical drives highly depends on the accuracy of the set voltage compared to its reference value, especially if the voltage is an input variable of integrated observers (e.g. flux observer). In general, voltage errors lead to a disturbance and, therefore, to a less-than-ideal control performance of the drive. Additionally, they increase harmonic currents within the motor and, hence, reduce its efficiency. Summarizing, there is a need for suitable inverter compensation schemes. Since it is difficult to generate and to parametrize a white-box inverter model, which precisely describes the inverter nonlinearity, in this paper, a black-box approach which utilizes machine learning is presented. Thereby, suitable neural network (NN) structures are selected in order to develop an appropriate inverter compensation scheme. Experimental results validate that trained neural networks can decrease the voltage errors by approx. 90 % compared to non-compensated control outputs and, at the same time, decrease the steady-state motor current harmonics significantly.

Vector-controlled induction machines (IM) are sensitive to incorrect model parameters and measurement errors when utilized as open-loop torque or speed-sensorless drives. Hence, there are manifold publications available addressing the optimization of the 1M's parameter identification as well as the flux- and speed-observer design. Up to now this was handled as a sequential two-step process. However, measurement uncertainty will adversely affect the 1M identification result and, consequently, the resulting uncertainty regarding the 1M's parameter will then adversely affect the observer design. The innovation of this contribution is to combine the observer design and machine identification by one global optimization step. As machine identification and observer tuning have the same goal, i.e. being able to control a given drive with highest speed and torque accuracy, there is no need to separate them from each other. By doing so, more degrees of freedom are given to the identification process. Moreover, since an extended Kalman filter is applied as flux- and speed-observer, sensor deviations are inherently considered due its working principle. Experimental tests proof the benefits of the proposed approach compared to the state of the art.

The permanent magnet synchronous motor (PMSM) is widely applied in highly utilised automotive traction drives and other industrial applications. With regards to the device life-time, thermal utilisation, safe operation and control performance, the permanent magnet temperature is of great interest. Since a direct magnet temperature measurement is not feasible due to constructional and cost issues in most applications, this contribution gives a review on state-of-the-art model-based magnet temperature determination techniques regarding PMSM. In this context, the existing publications can be classified into thermal models, flux observers and voltage signal injection approaches. Firstly, brief introductions of these methods are given, followed by a direct comparison regarding drawbacks and benefits. This evaluation exhibits the individual constraints of each approach motivating their fusion as an outlook of further investigations in this research field.

Monitoring rotor temperatures In permanent magnet machines is vital to avoid irreversible demagnetization. Different strategies have been developed among which the permanent magnet temperature observer (PMTO) stands out given its independence from signal injection and thermal models. In spite of its uncomplicated implementation, PMTO accuracy and sensitivity had been not deeply studied. This contribution presents a sensitivity analysis of the PMTO considering manufacture tolerances in the sensors that influence its performance: position, current, and DC-link voltage sensors. Results from Monte Carlo simulations showed a complex interaction between the sensor tolerances and achievable observer accuracy. While current sensor deviations play a minimum role in PMTO accuracy, voltage and position sensor deviations are critical. A statistical analysis reveled that a production quality control based on a sample of the population is not enough to guarantee an appropriate PMTO accuracy. Thus, specific sensor deviation ranges are required and were determined in this contribution.

High utilization of permanent magnet machines without monitoring their internal temperatures has negative impact on windings and permanent magnets. Lumped-parameter thermal networks (LPTNs) are therefore used to estimate magnet and winding temperatures. LPTNs identification is an intricate process as LPTNs can only be accurately described as linear-parameter varying systems (LPV). Thus specialized identification techniques are required such as global and local methods studied in the last decades. This paper studies the performance of the so-called glocal methods. Hence, SMILE, H 2 -norm, and H ∞ -norm methods are implemented and compared. All three glocal methods are able to represent the system with high accuracy. H 2 -norm and ∞ -norm methods achieve slightly better accuracy than SMILE; however, complications such as computational burden and local minimum convergence favor SMILE. The latter has a faster convergence and can achieve high accuracy with maximum temperature estimations errors of 6.8 °C, 6.2 °C, and 4.7 °C for the winding, end-winding, and permanent magnets. Finally, it was found that the model accuracy does not improve majorly by increasing the number of local models. It was estimated that a segmentation of the operating range (speed and current) into 4 or 5 parts respectively is enough to obtain a relative accurate LPV.

Monitoring critical temperatures in permanent magnet synchronous motors (PMSMs) is crucial to ensure safe operation and maximum device utilization as well. In this work, the application of recurrent neural networks featuring memory blocks (LSTMs/GRUs) are investigated upon their suitability to accurate temperature time series prediction inside PMSMs or similar motor types, which is the first time in literature to the author's best knowledge. Considered motor components are stator yoke, teeth and winding as well as the rotor's permanent magnets of a highly-utilized PMSM for electric vehicle applications. Having benchmark data available, numerous neural networks are trained and optimized with the aid of the Chainer framework and particle swarm optimization is conducted for finding suitable model hyper-parameters (e.g. number of hidden neurons or layers) on a computing cluster. It is found, that the Euclidean norm performance (in the range of 1–3 K) is similar but the worst-case predication errors (in the range of 9–14 K) are significantly higher compared to established modeling techniques like lumped-parameter thermal networks (LPTNs). This initial investigation motivates future research to increase ANN-based estimation accuracy by taken other ANN topologies, training methods or hyper-parameter optimization approaches into account.

Permanent magnet synchronous motors (PMSM) are widely used in many industry fields. Maximizing their degree of thermal utilization reduces the motor's weight, volume and cost. For this reason, real-time temperature information is required. However, this cannot be solely obtained from measurement sensors due to costs and constructional drawbacks. Thus, several direct temperature modeling techniques, like lumped-parameter thermal networks (LPTN), and indirect temperature estimation approaches based on accurate electrical motor models detecting temperature-sensitive parameter changes were investigated in the last years. As these methods are typically independent of each other, fusion approaches can be used to increase the conjoint estimation accuracy and robustness. In this contribution a Kalman filter (KF) is successfully combined with a low-order LPTN, an electrical model-based permanent magnet temperature observer and a built-in winding temperature sensor to achieve this goal. Measurement results for a 50 kW PMSM proof the enhanced KF performance compared to the individual estimation techniques.