To solve the problem that fault features are difficult to extract and the time-frequency features cannot fully represent the state information, a novel method is proposed in this paper based on the whale optimization algorithm (WOA) and the kernel extreme learning machine (KELM). First, the vibration signals are processed by the ensemble empirical mode decomposition and sample entropy to obtain the feature vectors. Based on this, a KELM model for fault diagnosis is established. Then, the penalty factor and the kernel parameters in the KELM are optimized by WOA to improve the stability and classification accuracy. Taking faults of a ball-screw pair on a linear feed table as a case, the experimental results indicate that the proposed method can effectively extract the fault features of the ball-screw pair, and it can achieve higher classification accuracy, faster convergence speed, and greater convergence precision than the existing fault diagnosis methods.

The vibration signals of rotating machinery usually contain various natural oscillation modes, exhibiting multi-scale features. This paper proposes a Multi-Branch one-dimensional deep Convolutional Neural Network model (MBCNN) that can extract multi-scale features from raw data hierarchically, thereby improving the diagnostic accuracy of gearbox faults in noisy environments. Meanwhile, the algorithms for multi-branch generation and algorithms of the convolution and pooling for each branch are deducted. The MBCNN integrates multiple branches with interrelated convolution kernels of different widths, and each branch can extract the high-level features of the signal. The network parameters of each branch are adjusted by the loss function, which makes the features of the branches complementary. Through the design of MBCNN, the local, global, deep layer and comprehensive information can be obtained from the raw data. On the widely used Case Western Reserve University Bearing Dataset, this paper conducted a performance comparison between the proposed MBCNN and other baselines including the shallow learning methods, 1D-CNN, and multi-scale feature learning methods. Moreover, our gearbox dataset was conducted on a fault diagnosis platform, and a series of experiments were conducted to verify the effectiveness and superiority of the MBCNN. The results indicate that the MBCNN can identify the faults in the gearbox with an accuracy of higher than 92%, and the average validation time per sample is less than 3.2 ms. In a noisy environment, the diagnostic accuracy can reach 90%. The proposed MBCNN provides an effective and intelligent detection method to identify the faults of rotating machinery in the manufacturing processes.

When using a ball-bar to measure rotary axis error, to keep the posture of the bar constant, a linkage motion of two linear axes with the rotary axis is required, which will cause the ball-bar length to contain both rotation axis error and linear axis error. In order to get the accurate rotary axis error, a decoupling strategy based on the actual position of the double balls on both ends of the ball-bar is proposed in a CFXYZA-type five-axis machine tool. Three measurement modes, axial, radial and tangential mode, are designed with a ball-bar instrument. Six groups of length variations of the ball-bar are measured by changing both the position and the height of the ball base mounted on the rotary table. Center offset error in measurement of the ball bar length is eliminated. The relationship between geometric error parameters of the rotary axis and ball-bar length variations is derived by the homogeneous coordinate matrix transformation from which the parameters can be identified. The displacement deviation aroused from the linkage motion of the linear axis with the rotary axis is also derived. Mathematical matching is operated between the feed rate of the rotary axis and acquisition speed of the ball-bar software. Center offset of the ball-bar length data acquired is eliminated to reduce the influence of installation. An obvious scaling mismatch between the axes diagnosed from the polar plot of ball-bar length variation shows that the C -axis has the tilt error around the X - or Y -axes. Experiment results show that the proposed method can improve the identification accuracy. The decoupling method is both effective and convenient for identifying geometric error of the rotary axis in a five-axis machine tool.

A proper hardening depth is critical to the load-bearing capacity of a part, and heat treatment, including carburizing and quenching, can highly determine the hardness distribution in the part’s surface after manufacturing. This paper proposes a ‘hardness prediction and parameter optimization’ approach that deploys the finite element method (FEM), the artificial neural network (ANN), and the Genetic Algorithm (GA), to describe the relationships between the carburizing/quenching parameters and the hardening depths and conversely to determine the optimized parameters for a given hardening depth. First, the numerical models for carburizing, quenching, and the hardness field are built respectively. And based on these models, the finite element simulation model is designed to predict the carbon content, the microstructure and the hardness of the part. A BP network is then trained by using the data obtained from the finite element simulation, and the model between the carburizing/quenching parameters and the hardening depths on part is established. The optimization model for the carburizing/quenching parameters is finally established through GA, which can determine the optimized parameters for a given hardening depth. The effectiveness of the ‘prediction-optimization’ approach is verified by a series of experiments. The hardening depth predicted by the proposed approach holds a 10% relative error from that measured in the carburizing and quenching experiment. And the optimized parameters for the heat treatment process can work as a meaningful reference for the heat treatment.