The stator flux linkages serve as a key to the optimal control of synchronous machines (SMs). However, due to their complex and nonlinear characteristics, accurately modeling and identifying them online remains highly challenging. In this regard, neural network-based learning strategies are considered promising candidates for modeling the flux linkages, but their application has so far been largely limited to offline training of neural networks. Therefore, this study presents a physics-informed online learning method for accurately modeling the flux linkages of SMs. The proposed method enables online training of a neural network to learn the physical laws governing the flux linkages while adhering to the model’s inherent physical constraints. The learning rules for updating the neural network weights are formulated to satisfy the first-order optimality conditions, and the proposed method can be employed as an online flux linkage estimator. The effectiveness of the proposed method is validated through simulation results conducted on a 35 kW interior permanent magnet synchronous machine (IPMSM) drive.

This paper proposes a design method for linear matrix inequality (LMI)-based neural network (NN) observer gain in discrete-time domain. The proposed scheme employs an NN with a single hidden layer to approximate the lumped nonlinear term which includes uncertainties. A Lyapunov function is constructed to guarantee the stability of both the linear observer and the NN updates. The observer gain is determined by solving the LMI conditions, and the design is simplified by minimizing the number of tuning parameters, using a common gain structure for all vertices. Furthermore, designing an H∞observer can reduce the effect of NN approximation error and the measurement noise.The key advantages of the proposed method lie in its optimal LMI-based observer gain design, minimal tuning parameter requirement, and the capability to estimate both the system states and the lumped nonlinear term simultaneously. Simulation results indicate that the proposed method successfully tracks the actual states and the lumped nonlinear term and reduce the effects of NN approximation error and the measurement noise with comparison of the root mean square error (RMSE) values.

This paper proposes an online actor critic learning-based optimal tracking control method for output-feedback servo positioning systems under unknown external disturbances. The servo system is reformulated into a control-affine form, where the uncertain dynamics are compactly represented as a lumped unknown function. An online identifying filter is introduced to estimate these dynamics, while an actor-critic neural network structure is used to approximate the value function and optimal control input. The proposed method yields an approximate solution to the Hamilton-Jacobi-Bellman equation, with adaptive update laws ensuring asymptotic convergence of the Bellman residual error. Lyapunov-based analysis guarantees the stability of the closed-loop system. Simulation results confirm the effectiveness of the proposed method in achieving robust tracking under time-varying disturbances.

This paper presents a constrained optimization-based neuro-adaptive control (CONAC) for nonlinear synchronous machines (SMs) under voltage constraints, which allows controlling the completely unknown electrical drive system, after a brief learning phase with very satisfactory control performance. The artificial neural network (ANN) in the proposed neuro-adaptive controller (NAC) learns online and empowers the controller to handle parameter uncertainties. Moreover, it solves a constrained optimization problem which allows considering the nonlinear voltage constraints as well, by deriving the adaptation laws of the ANN’s weights from the Lagrangian function. Boundedness of tracking error, convergence of the ANN weights, and satisfaction of the constraints are guaranteed by Lyapunov theory. Numerical simulations in combination with a realistic (nonlinear) synchronous machine drive demonstrate the effectiveness and robustness against parameter and modeling uncertainties of the proposed NAC and its very acceptable constraints handling.

The predictive energy management (PEM) of hybrid electric vehicles (HEVs) is a challenging problem of trajectory optimization involving future information. Most previous studies have presented intuitive methods for selecting parameters that represent future information. However, such intuitive methods lack theoretical analysis and do not guarantee global optimality. This study adopts the novel perspective that the PEM problem can be analytically solved so as to ensure near-optimal efficiency. The key idea is to reformulate the trajectory optimization problem as a quadratic programming (QP) problem based on optimal control principles. The equivalence between the original and QP problems is theoretically derived. The proposed PEM strategy is then implemented by solving the QP problem at every sampling time to obtain the optimal control input. Whereas the original problem requires information on the entire future trajectory, which is difficult to predict accurately, the QP problem requires only lumped parameters, specifically, the energy demands and time durations of each segment of the future path. These lumped future driving parameters can potentially be predicted using information obtained through vehicle connectivity. Simulation results obtained under a real-world driving scenario show that the proposed PEM strategy provides a control result in real time (within 2 ms) that is very close to the globally optimal solution both qualitatively and quantitatively, with a loss of optimality of only 0.14%.

Distributing workloads that cannot be handled by a single edge device across multiple edge devices is a promising solution that minimizes the inference latency of deep learning applications by exploiting model parallelism. Several prior solutions have been proposed to partition target models efficiently, but most studies have focused on finding the optimal fused layer configurations, which minimize the data-transfer overhead between layers. However, as recent deep learning network models have become more complex and the ability to deploy them quickly has become a key challenge, the search for the best fused layer configurations of target models has become a major requirement. To solve this problem, we propose a lightweight model partitioning framework called Legion to find the optimal fused layer configurations with minimal profiling execution trials. By finding the optimal configurations using cost matrix construction and wild card selection, the experimental results showed that Legion achieved a similar performance to the full configuration search at a fraction of the search time. Moreover, Legion performed effectively even on a group of heterogeneous target devices by introducing a per-device cost-related matrix construction. With three popular networks, Legion shows only 3.4% performance loss as compared to a full searching scheme (FSS), on various different device configurations consisting of up to six heterogeneous devices, and minimizes the profiling overhead by 48.7× on average.

Graphics processing units (GPUs) have been commonly utilized to accelerate multiple emerging applications, such as big data processing and machine learning. While GPUs are proven to be effective, approximate computing, to trade off performance with accuracy, is one of the most common solutions for further performance improvement. Precision scaling of originally high-precision values into lower-precision values has recently been the most widely used GPU-side approximation technique, including hardware-level half-precision support. Although several approaches to find optimal mixed-precision configuration of GPU-side kernels have been introduced, total program performance gain is often low because total execution time is the combination of data transfer, type conversion, and kernel execution. As a result, kernel-level scaling may incur high type-conversion overhead of the kernel input/output data. To address this problem, this paper proposes an automatic precision scaling framework called PreScaler that maximizes the program performance at the memory object level by considering whole OpenCL program flows. The main difficulty is that the best configuration cannot be easily predicted due to various application- and system-specific characteristics. PreScaler solves this problem using search space minimization and decision-tree-based search processes. First, it minimizes the number of test configurations based on the information from system inspection and dynamic profiling. Then, it finds the best memory-object level mixed-precision configuration using a decision-tree-based search. PreScaler achieves an average performance gain of 1.33x over the baseline while maintaining the target output quality level.

Electrotactile stimulation has been used to provide sensory information of forearm prosthesis to users. Although conventional sensory feedback method, where one electrode expresses sensory information of only one finger, could provide force information of three fingers by using three electrodes, it showed less cognitive accuracy when more than two electrodes were stimulated simultaneously compared to individual stimulation. To improve the cognitive accuracy, we presented a sensory feedback method called two-channel electrotactile stimulation for the thumb, index and middle fingers using only two electrodes. The force information of the index and middle fingers was delivered into two electrodes, respectively, by intermittent stimulation. The information of the thumb was delivered to the user by inserting additional offset pulses onto the channel of the index finger based on an assumption that the force of thumb is proportional to the sum of the forces of the index and middle fingers. The presence of offset pulses in the channel indicates the binary state of thumb if it contacts with an object or not. We conducted two psychophysical experiments where healthy subjects classified the binary states of each finger and identified intensity from two-channel stimuli. The cognitive accuracies for intensity identification were 78.8% and 62.2% for the intermittent stimulation and conventional method, respectively, and accuracy for classifying the fingers was 93.1% for every combination of the three fingers. The results demonstrated that the proposed two-channel electrotactile stimulation could be an attractive method to express the information of fingers of prosthesis with high accuracy.