Fault Diagnosis on Induction Motors Using

A scheme for diagnosis and identification of mechanical unbalances and shaft misalignment on machines driven by induction motors is presented in this work. Fault identification is performed using unsupervised artificial neural networks: the so-called Self-Organizing Maps (SOM). The information of the motor phase current is used for feeding the network, in order to perform the fault diagnosis. The network is trained using data generated through the simulation of a motor- load system model. Such model allows including the effects of load unbalance and shaft misalignment. Experimental data are later applied to the SOM in order to validate the proposal. It is demonstrated that the strategy is able to correctly identify both unbalanced and misaligned cases. Predictive maintenance attempts to avoid unexpected faults in the industry, which cause great economic losses due to interruptions in continuous production processes. Hence arises the need and interest for the industry to develop strategies for on-line detection and diagnosis of incipient faults in electrical machines. In this way, process interruptions can be planned and machines maintenance can be performed during programmed stops. This allows reducing the maintenance time and the associated economic losses. Among these strategies, those based on measurement of motor voltages and currents allows detecting different types of faults by measuring from the switchboard, thus reducing the risks for the operator in hazardous environments or difficult to access. Such strategies have also been used for detecting problems associated to the load driven by the motor. The detection and diagnosis of electrical or mechanical faults on induction motor implies, in most cases, the interpretation of the frequency spectrum of the motor current, power, Park's vector, among others (1). This requires an expert who performs the task, based on the information obtained from the processed signals. At present the study of different alternatives, such as Artificial Intelligence (AI) techniques, have taken great importance because they require a minimal interpretation of the studied system. Thus, the diagnosis task is simplified (2). Unsupervised Neural Networks (UNN) was proposed for fault diagnosis on electric drives in (3). The application of UNN for the diagnosis of load unbalances and shaft misalignment problems in electric drives is presented in this work. A Self-Organizing Map (SOM) - a type of UNN - is implemented and trained using simulation data obtained from a motor-load model, which allows considering load unbalance and shaft misalignment. This network is later used to obtain an automatic diagnosis of such problems, from data obtained from measurements made on real cases.