Ball Screw System Condition Monitoring And Performance Evaluation Based On Representation Learning

Ball screw, a mechanical actuator which converts rotation to linear motion, is widely used in precise mechanical positioning and measurement system. It will inevitably appear wear, loose structure and some other performance degradation during operation. The failure or performance degradation of ball screw system will seriously affect the accuracy and safety of mechanical equipment. So it is important to research condition monitoring and performance evaluation of ball screw to improve manufacturing standard and reduce the maintenance cost. The vibration signal of ball screw contains significant condition status and performance information. However the acquired vibration signal is usually merged in noise that makes it difficut to extract the useful information. Meanwhile, how to efficiently extract and express useful information in signal is a hot and difficult research topic in signal processing field. In this paper, we did deep research about ball screw condition monitoring and performance evaluation based on representation learning. Below are the details.(1)Study on the basic structure and learning algorithm of representation learning, especially dictionary learning and deep learning. The essential attribute of representation learning, learning representation of the data that makes it easier to extract useful information without the help from prior knowledge, is revealed in the paper. In the meantime, two image processing experiments are conducted to indict the effectiveness and capacity of representation learning.(2)Study on mechanical vibration signal denoising technique based on dictionary learning and sparse coding. According to the characteristics of ball screw axial bearing vibration, the complex heterogeneous training sample set is constructed based on some fixed dictionary functions and bearing vibration signal. Explore the denoising method based on the sparse representation in the learned dictionary region. And the online dictionary learning algorithm is applied to reduce the time cost. The experiments based on simulation and acquisition data demonstrated the effectiveness of the proposed method.(3) Design a ball screw system fault locating algorithm under the variable speed conditions. The polynomial Chirplet transformation is adopted to calculate the instantaneous frequency of screw rotation. An anomaly detection algorithm based on amplitude threshold and the derivative threshold is proposed to detect the fault time. Then the true fault location is calculated. Three different varying speed experiments are conducted to test the proposed method. The results show that the proposed method can locate the real fault, and the error is less than two screw lead.(4) A ball screw accelerated performance degradation testbed is designed. According to the life formula of ball screw, an experiment scheme is estimated. And some different operation conditions are simulated with varying of speed and load. And the signal of vibration and torque is acquired, that is the basic data set for testing the performance evaluation algorithm.(5) A performance evaluation method based on deep neural network is proposed. The time domain, frequency domain and time-frequency domain features are extracted to represent the characteristics of ball screw vibration signal. And a new frequency domain feature that is the correlation coefficient of frequency kurtosis is proposed. The deep neural network is adopted to learn the high-level feature and reduce the dimension to achieve the accuracy. The denoising deep learning structure, which substitutes the specific input location value with ransom numer, is applied to improve the robustness.(6) The practical technique of ball screw condition monitoring and performance evaluation is studied based on the above-mentioned theories and experiments. The remote operation is implemented based on the communication protocol with Socket. And the features extraction and model calculation are accelerated with the help of GPU. Then the GUI is completed through the WxPython that is helpful for convenient input and result display.In this paper, the ball screw performance degradation experiment, data representation learning, condition monitoring and performance evaluation are deeply studied. Explore the new vibration signal processing method based on representation learning, and a new adaptive vibration signal representation scheme is proposed. The practical study of ball screw condition monitoring and performance evaluation shows positive effect in industrialization of the system.