School of Automobile and Traffic Engineering, Wuhan University of Science and Technology, Wuhan 430081, China.
Department of Mechanical Engineering, School of Engineering, Cardiff University, Cardiff CF24 3AA, UK.
Sensors (Basel). 2022 Sep 1;22(17):6628. doi: 10.3390/s22176628.
Machine tools, as an indispensable equipment in the manufacturing industry, are widely used in industrial production. The harsh and complex working environment can easily cause the failure of machine tools during operation, and there is an urgent requirement to improve the fault diagnosis ability of machine tools. Through the identification of the operating state (OS) of the machine tools, defining the time point of machine tool failure and the working energy-consuming unit can be assessed. In this way, the fault diagnosis time of the machine tool is shortened and the fault diagnosis ability is improved. Aiming at the problems of low recognition accuracy, slow convergence speed and weak generalization ability of traditional OS recognition methods, a deep learning method based on data-driven machine tool OS recognition is proposed. Various power data (such as signals or images) of CNC machine tools can be used to recognize the OS of the machine tool, followed by an intuitive judgement regarding whether the energy-consuming units included in the OS are faulty. First, the power data are collected, and the data are preprocessed by noise reduction and cropping using the data preprocessing method of wavelet transform (WT). Then, an AlexNet Convolutional Neural Network (ACNN) is built to identify the OS of the machine tool. In addition, a parameter adaptive adjustment mechanism of the ACNN is studied to improve identification performance. Finally, a case study is presented to verify the effectiveness of the proposed approach. To illustrate the superiority of this method, the approach was compared with traditional classification methods, and the results reveal the superiority in the recognition accuracy and computing speed of this AI technology. Moreover, the technique uses power data as a dataset, and also demonstrates good progress in portability and anti-interference.
机床作为制造业中不可或缺的设备，广泛应用于工业生产中。恶劣而复杂的工作环境容易导致机床在运行过程中发生故障，因此迫切需要提高机床的故障诊断能力。通过识别机床的运行状态（OS），可以确定机床故障的时间点和耗能单元，从而缩短机床的故障诊断时间，提高故障诊断能力。针对传统 OS 识别方法存在识别准确率低、收敛速度慢、泛化能力弱等问题，提出了一种基于数据驱动的机床 OS 识别的深度学习方法。该方法可以利用数控机床的各种功率数据（如信号或图像）识别机床的 OS，然后直观地判断 OS 中包含的耗能单元是否出现故障。首先，采集功率数据，使用小波变换（WT）的数据预处理方法对数据进行降噪和裁剪预处理。然后，构建 AlexNet 卷积神经网络（ACNN）来识别机床的 OS。此外，研究了 ACNN 的参数自适应调整机制，以提高识别性能。最后，通过案例研究验证了所提方法的有效性。为了说明该方法的优越性，将其与传统分类方法进行了比较，结果表明，该 AI 技术在识别准确率和计算速度方面具有优越性。此外，该技术使用功率数据作为数据集，在可移植性和抗干扰性方面也取得了良好的进展。
