Application of Hidden Markov Model for Fault Analysis of ODO in Train Location Unit

During the running of a train, the lock-and-slip state of a wheel may occur and hence, an Odometers (ODO) will temporarily display the state of functional failure. To address this issue, fault analysis of ODO based on the application of hidden Markov model (HMM) has been carried out. Firstly, an algorithm based on neural networks (NNs) was applied for the fault diagnosis of ODO. A fault diagnosis method based on HMM was then studied from the perspective of pattern recognition. The observation of the quantization coding sequence was introduced into the HMM classifier for establishing the hidden fault states of the ODO model with the aid of training data. By entering the observation sequence matching with the HMM classifier, the effective state of ODO was obtained. Finally, genetic algorithms (GAs) were employed for obtaining the parameters of HMM, rather than employing the Baum-Welch (B-W) algorithm. The obtained case study results show that the fault identification accuracy based on NN is 33.3%, the diagnos-tic accuracy based on HMM can reach up to 100% between normal and fault conditions, and the overall diagnostic accuracy is 95%. The training speed of GA can help in reaching the steady state quickly and the training accuracy can be improved by 86%. Furthermore, the obtained results show that the classification ability of NN is comparatively weak and cannot meet the required needs of classification when there is high noise in the data. The GA can improve on the drawbacks of the B-W algorithm that in turn can easily fall into the local optimum. The fault diagnosis method based on HMM has higher accuracy when compared with the NN-based method.