Recent diagnostic approaches based on the deep learning model have attracted much attention. However, developing an outstanding AI diagnostic model requires many training samples with labeled information. Moreover, training deep models is labor-intensive and time-consuming, and labeling samples and training models increase workload. To overcome these problems, this article proposes an unsupervised deep transfer learning (DTL) method with an isolation forest (iForest) for machine fault diagnosis. First, the isolation forest is used to classify and label the samples automatically; then, these labeled data are used to train deep learning (DL) models; finally, small data with the label of the target domain are used to fine-tune parameters and complete the fault diagnosis. The proposed approach has been validated with the fan gearbox dataset, the bearing dataset, and the ball screw dataset. The results show that the proposed unsupervised deep transfer learning model has high accuracy and generality.

In robotic arms, transmission error will cause positioning error, which increases with the wear and tear of the transmission mechanism. Positioning error will degrade the performance of the robotic arm. It is necessary to carry out intelligent maintenance for the robotic arm. In this paper, a low-cost digital twin-driven positioning error compensation method is proposed for robotic arm. The digital twin model of the robotic arm is first established, including the physical layer, data transfer layer, model layer, functional layer, and application service layer. A low-cost attitude sensor is then applied to provide actual information of the robotic arm to the digital twin model. When the digital twin model senses the positioning error of the robotic arm, it reduces the positioning error by adjusting the motor angle. The effectiveness of the proposed method is finally verified by the experimental results under different conditions.

Robotic arms for industrial applications rely on expensive, complex solutions for high-precision positioning error compensation. Digital twins provide virtual representations of physical assets to optimize their engineering performance, which helps address the above issues. To address this problem, this paper proposes a digital twin-driven 3D position information mutuality and positioning error compensation for robotic arm. A digital twin model is developed and a virtual sensor is modeled geometrically. Information exchange between the physical and virtual sensor enables comparison of the actual and target arm pose. Through closed-loop alignment of the physical sensor data with the virtual output, the arm joints are dynamically adjusted to reduce positioning errors. Information mutuality significantly reduces the amount of calculation necessary to determine the robotic arm’s actual angle of motion. Experimental results from various positions have confirmed the efficacy of the method, with a remarkable 81.23% reduction in positioning error.

Recently, random forest (RF) as a highly flexible machine learning algorithm has been applied to medicine, biology, machine learning, computer vision and other fields, and has shown good application performance. Nevertheless, the operation efficiency and identification accuracy of RF algorithm are actually affected by the number of the decision trees. A novel RF model, referred to as the extreme random forest (ERF), was proposed to improve the ability of feature extraction and reduce the computation burden. In the ERF method, the dimensionality of the high-dimensional data is randomly reduced through the random mapping matrix, and the classification performance after dimensionality reduction is improved. In this way, the sample dimension of the input RF is greatly reduced, which improves the operation efficiency of the RF. Both theoretical analysis and experiment tests have verified the superiority of the proposed method. In the experimental part, the present ERF method was compared with other peer method in terms of diagnostic performance and computational efficiency. The comparison results showed that the ERF method has more advantages both in diagnostic accuracy and computational efficiency. In addition to mechanical fault diagnosis, the proposed ERF can also be used in other machine learning fields.

Low torque ripple level is the prerequisite for surface-mounted permanent magnet synchronous machines (SPMSM) to maintain control accuracy and speed stability. Since the speed harmonic's amplitude is proportional to the torque harmonic's amplitude, torque ripple can be reduced by speed harmonic minimization. Harmonic injection method is a practical speed harmonic minimization method. But current harmonic regulation and complex theoretical derivation are required in existing methods. What's more, the effectiveness of many methods will deteriorate when the parameters of PMSM and PI controllers vary. In order to avoid the influence of parameter variations and complex algorithms, an amplitude-phase based optimal voltage harmonic injection method is proposed for speed harmonic minimization in this paper. The influence of the injected voltage harmonic's amplitude and phase on the speed harmonic's amplitude is firstly analyzed. And then a simple speed harmonic controller based on the optimization of injection voltage harmonic's amplitude and phase is designed. The proposed method does not rely on the motor parameters. Moreover, the proposed method is easy to implement. It does not need to regulate current harmonics. The proposed method is simulated and experimentally evaluated under both steady-state and transient conditions.

Unbalanced data set is a common problem in the practical application of deep learning algorithms, and it often limits the diagnostic ability of identification models. Therefore, a sample enhancement method combining generative adversarial network (GAN) and feature moving is proposed for solar panel defect detection, thereby improving the defect recognition accuracy. This method is mainly divided into the following three steps: firstly, the traditional GAN network is used to generate samples to reduce the problem of data imbalance; then, the feature moving method is proposed to be integrated into the YOLOv5S model to reduce the distribution difference between the generated samples and the actual samples, which further improving the sample quality and enhance the accuracy of defect identification; finally, the proposed method is effectively verified by three types of unbalanced solar panel defect data (over-etching, fingerprint, and scratch). When using GAN to generate samples, the results of the ablation and comparison experiment show that the single-class average precision (AP) can be increased by up to 2.8% and the mean average precision (mAP) can be increased by 1%. When using styleGAN3 to generate samples, AP and mAP can be improved by 0.2%.