In the past years, traditional pattern recognition methods have made great progress. However, these methods rely heavily on manual feature extraction, which may hinder the generalization model performance. With the increasing popularity and success of deep learning methods, using these techniques to recognize human actions in mobile and wearable computing scenarios has attracted widespread attention. In this paper, a deep neural network that combines convolutional layers with long short-term memory (LSTM) was proposed. This model could extract activity features automatically and classify them with a few model parameters. LSTM is a variant of the recurrent neural network (RNN), which is more suitable for processing temporal sequences. In the proposed architecture, the raw data collected by mobile sensors was fed into a two-layer LSTM followed by convolutional layers. In addition, a global average pooling layer (GAP) was applied to replace the fully connected layer after convolution for reducing model parameters. Moreover, a batch normalization layer (BN) was added after the GAP layer to speed up the convergence, and obvious results were achieved. The model performance was evaluated on three public datasets (UCI, WISDM, and OPPORTUNITY). Finally, the overall accuracy of the model in the UCI-HAR dataset is 95.78%, in the WISDM dataset is 95.85%, and in the OPPORTUNITY dataset is 92.63%. The results show that the proposed model has higher robustness and better activity detection capability than some of the reported results. It can not only adaptively extract activity features, but also has fewer parameters and higher accuracy.

The commutation torque ripple of a brushless dc motor (BLDCM) in a six-step driving mode generates vibration noise. To minimize the commutation torque ripple of the BLDCM, this article analyzes the cause of torque ripple and establishes the model of torque ripple suppression by using the phase current predictive method based on the trapezoidal back electromotive force (EMF). In addition, based on the prediction model of the non-commutation phase current, the pulsewidth modulation model predictive control (PWM-MPC) algorithm is proposed. By changing the duty cycle with prior evaluation, the predictive control avoids commutation current hopping, thereby reducing torque ripple during commutation. This control approach is more accessible to be implemented, because it does not require changing the topology of the motor driving circuit. Simulation models of the proposed control scheme constructed in the MATLAB/Simulink environment are given, compared with the conventional square-wave driving method. Moreover, experiments are performed to verify the feasibility. The output torque during experiments is transmitted to the computer software through the torque transducer. Compared with the traditional driving methods, the simulation and experimental results show that the proposed novel algorithm in this article could suppress the torque ripple efficiently.

Direct current (DC) serial arc faults usually occur in the damaged insulation lines or line connections, which will cause serious accidents such as fires and explosions. With the rapid increase of electric vehicles, DC serial arc faults are more and more dangerous to battery system. Therefore, a binary classification model based on machine learning algorithm was proposed to detect DC serial arc faults effectively in this study. It was optimised according to the characteristic signals of the arc to be satisfied with different loads for higher detection accuracy and robustness. In the simulative experiments for the power system electric vehicle, while the loads changing to the motor, the resistor or the inverter, it will all reach a highly successful detection rate, respectively.

Electric vehicle (EV) power system is flammable and explosive when the direct current (DC) arc occurs at elevated temperature. Thus, DC serial arc real‐time monitoring is an insurance to keep away from disaster. In this study, the detection algorithm of DC serial arc is proposed. The wavelet entropy algorithm, the classification model based on support vector machine and logistic regression are analysed separately. The above algorithms are combined to identify the DC serial arc faults effectively under different types of loads in EV power system. The results show that the combined algorithm has a good performance of DC serial arc detection with high accuracy and robustness compared with a simple approach. Meanwhile, the false detection rate of the detection algorithm is close to zero, which could ensure the safety and stable operation of the system.

Traditional torque measurement has greatly depended on the contact between sensors and the rotating shaft. However, the high-speed rotation of the shaft leads to significant sensor wear, compromising the stability of the measurement system. Additionally, the newly collected test data tend to be high in latency and noise, which introduces accuracy challenges for artificial intelligence algorithms in torque measurement. This paper presents the measurement method, hardware platform design, data calibration and software optimization of a torque and rotational speed measurement system, which uses a non-invasive method and accurately estimate torque and rotational speed under both low and high-speed operating conditions by analyzing the mechanical phase difference between two pulse signals. The pulse signals are precisely captured by Hall sensors and an FPGA microcontroller circuit, while signal delay and noise are optimized through a CNN-LSTM artificial intelligence algorithm and system calibration programming. To verify the correctness of the proposed method, a test platform equipped with Hall sensors and servo motor and 200N·m standard torsion device for measuring the torque and rotational speed under torque force generation device's control was implemented. The effectiveness of the proposed method is demonstrated by achieving a measurement error rate within 1%, with the rotational speed error reduced from 3.15% to 0.86% and the torque error decreased to 0.96%.