Abstract: To meet the failure diagnosis requirements of complex Computer Numerical Control (CNC) machine,an Adaptive Error Correction Probabilistic Neural Network (AECPNN) method by using the advantages of artificial neural network in the field of failure diagnosis was put forward to diagnose real-time failure of CNC machine.Aiming at the problem that the biased judgment interface leaded by Probabilistic Neural Network (PNN) which didnt consider the crisscross influence among different patterns,an adaptive error correction model was proposed based on PNN.Through adaptive clustering and batch of correction for same categorys false classified samples,the judge interface was re-planned.Experimental results of two-spirals-apart problem and IRIS problem showed that AECPNN method was superior to PNN and Radial Basis Probabilistic Neural Network (RBPNN) in aspects of classification accuracy rate and model generalization ability.In addition,the training and testing speeds of AECPNN could meet real-time demand of classification.Applications in the failure diagnosis of CNC machine demonstrated that the proposed method was feasible and effective due to high accuracy rate of failure pattern recognition.

Key words: failure diagnosis, probabilistic neural network, adaptive error correction, computer numerical control machine

摘要： 针对数控机床自身复杂性对故障诊断模型的需求,借助人工神经网络在故障诊断领域的优势,提出一种基于自适应误差修正模型的概率神经网络,以实现数控机床机械故障的实时诊断。针对概率神经网络由于未考虑模式间交错影响而导致判决界面有偏的问题,在概率神经网络的基础上设计自适应误差修正模型,通过对同类别错误分类样本进行自适应聚类并批量修正的过程,实现判决界面的重新规划。对双螺旋分类问题、IRIS分类问题的实验结果表明,该方法在分类准确率和模型泛化能力方面均优于概率神经网络方法和径向基概率神经网络方法,且训练速度和测试速度满足分类实时性需求。在数控机床故障诊断领域的应用表明,所提方法的故障模式识别准确率高,具有可行性和有效性。

关键词: 故障诊断, 概率神经网络, 自适应误差修正, 数控机床