Bearing fault diagnosis holds immense significance for maintaining the stable operation of bearings. However, existing approaches encounter formidable challenges regarding accuracy and efficiency. This paper proposes an innovative approach that leverages a deep - learning model optimized by the improved Whale Optimization Algorithm (IWOA), enabling accurate and efficient bearing fault diagnosis. In terms of diagnosis model construction, this paper integrates the multi-scale Convolution Neural Network (MSCNN), Bidirectional Mogrifier-Gated Recurrent Unit (BiMGRU), and Multi-head Attention Mechanism (MHAM) to develop the MSCNN-BiMGRU-MHAM model. MSCNN serves to extract multi-scale spatial features, enhancing the model's ability to learn the characteristics of diverse data and improving the model generalization under variable working conditions. BiMGRU captures the temporal correlation features, effectively enhancing the context information interaction. MHAM performs parallel calculations, achieving parallel processing of fault-related information and improving the diagnosis efficiency. What’s more, for model training, an IWOA, integrating a chaotic strategy, a convergence factor nonlinear strategy, and a weight adaptive strategy, is proposed. It is utilized to optimize the model's crucial hyper-parameters synchronously, achieving to the optimal structure. This approach overcomes the drawbacks of relying on manual experience for parameter adjustment in traditional models. Consequently, significantly improving the efficiency and accuracy of model training. The effectiveness of the proposed approach is validated through bearing fault diagnosis experiments under noisy and diverse working conditions. Both public and real datasets are employed in these experiments. The results demonstrate that the proposed approach outperforms existing methods, underscoring its substantial practical application value.

In this work, a new method based on the nonlinear output frequency response functions (NOFRFs) and deep convolution neural network (CNN) is proposed for RV reducer fault diagnosis. In order to solve the problem of low accuracy caused by output signal that cannot describe the nonlinear characteristics of system, the nonlinear spectrum based on output frequency response function is adopted. To improve the performance of error cost function model in back propagation (BP) process of CNN, a new comprehensive error cost function model is designed to guide the network parameters' optimization in direction of the feature classification. Besides, an adaptive network structure optimization algorithm is proposed to avoid the blindness of network structure selection by experience. Specifically, 4-order NOFRF spectrums of system are first obtained by adaptive identification algorithm. Next, NOFRF spectrums are transformed into 2-D images, and CNN extracts the fault features hidden in spectrum images and performs fault classification. The experiment results indicate that the proposed method is effective for RV reducer fault diagnosis with the accuracy of 99.69%.

An existing problem in the robotic field is to solve the inverse kinematics (IK) problem of redundant robot with high speed and high precision. A novel IK optimization method based on the Gaussian Damped Least Squares (GDLS) is proposed in this paper. A significant contribution of this method is to make the iteration converge in a faster and more accurate way by introducing an optimal enhanced step-size coefficient. The machine learning model can be trained with 10 ⁶ data points in the reachable region of the robot, and the optimal enhanced step-size coefficient in each solving process can be predicted by the model. The accuracy and stability of the algorithm proposed are verified through an example of an arbitrary 7R redundant robot. The average number of iterations is less than 10, with super high solving speed. Furthermore, the algorithm also has better convergence, which can reach 96.23% when the error threshold is 0.01 mm. The common IK methods are evaluated in this paper, and the results show that the optimized method has good performance in convergence, accuracy and speed.

To solve the problem of low accuracy in traditional fault diagnosis methods, a novel method of combining generalized frequency response function(GFRF) and convolutional neural network(CNN) is proposed. In order to accurately characterize system state information, this paper proposed a variable step size least mean square (VSSLMS) adaptive algorithm to calculate the second-order GFRF spectrum values under normal and fault states; In order to improve the ability of fault feature extraction, a convolution neural network (CNN) with gradient descent learning rate and alternate convolution layer and pooling layer is designed to extract the fault features from GFRF spectrum. In the proposed method, the second-order GFRF spectrum of each state of Permanent Magnet Synchronous Motor (PMSM) is obtained by VSSLMS; Then, the two-dimension GFRF spectrum, which is regarded as the gray value of the image,will be further transformed into image. Finally, the CNN is trained with learning rate by gradient descent way to realize the fault diagnosis of PMSM. Experimental results indicate that the accuracy of proposed method is 98.75%, which verifies the reliability of the proposed method in application of PMSM fault diagnosis.