Gearbox is a significant part for the transmission of vehicles and various mechanical devices and is being utilized broadly in the industries despite of its failure prone nature. Therefore, the need arises for diagnosing the faults present in a gearbox and to rectify the faulty gear. In this paper, deep learning method is utilized for the diagnosis of faulty gears and employs the modified AlexNet for the classification of various gear signals. The hidden units present in the bidirectional LSTM (long short term memory) layer of the AlexNet is selected by proposing an improved grasshopper optimization algorithm (IGOA). After the process of classification, performance evaluation is carried out for various performance measures. It is found that proposed method achieves accuracy of 2.4 %, specificity of −0.3 %, sensitivity of 1.01 %, recall of 0.97 %, precision of 0.59 %. Based on the results obtained it is found that proposed algorithm is more efficient when compared to existing algorithm.

变速箱是车辆和各种机械设备的变速箱的重要组成部分，尽管其容易发生故障，但仍在工业中得到广泛使用。因此，需要诊断齿轮箱中存在的故障并纠正有故障的齿轮。本文将深度学习方法用于齿轮故障的诊断，并将改进的AlexNet用于各种齿轮信号的分类。通过提出一种改进的蚱hopper优化算法（IGOA），可以选择AlexNet的双向LSTM（长期短期记忆）层中存在的隐藏单元。在分类过程之后，对各种绩效指标进行绩效评估。结果发现，提出的方法的准确度为2.4％，特异性为-0.3％，灵敏度为1.01％，召回率为0.97％，精度为0.59％。根据获得的结果，发现与现有算法相比，提出的算法效率更高。