Fault diagnosis of rotating machines is one of the most considered maintenance methods for detecting faults early to save maintenance cost and time. In this work, an improvement technique is presented using back propagation neural network (BPNN) based vibration data to detect different faults in rotating machines such as unbalance, pulley misalignment, belt damage, and combined faults. The root means square (RMS) of vibration signals at different points was collected and employed as an input vector to the network. It was observed that the test and validation performance achieve the same pattern and the best validation was recorded at 0.33038 mean squared error (MSE). This training accuracy can identify combined pulley misalignment with unbalance, static unbalance on two shafts, dynamic unbalance, and combined belt damage with unbalance faults with identification accuracy of 95, 92, 88, and 80%, respectively. Static unbalance, pulley misalignment, and belt damage faults come in the second level of accuracy since they have the same accuracy of 75%. Furthermore, this network has a superior improvement in detecting combined faults in addition to other single variable faults. |
