Vibration analysis is an established method for fault detection and diagnosis of rolling element bearings. However, it is an expert oriented exercise. To relieve the experts, use of Artificial Intelligence (AI) techniques such as deep neural networks, especially convolutional neural networks (CNN) have gained much attention of the researchers because of their image classification and recognition capability. Most of the researchers convert the vibration signal into representative time frequency vibration images such as spectrogram, and scalogram. These images are used as input to train the CNN model for fault diagnosis. Commonly, fault diagnosis is performed under same operating conditions, where models are trained and deployed for prediction under same operating conditions. However, outside laboratory environment, in real world applications, different operating conditions, such as variable speed, may be encountered. With the change in speed, characteristic frequencies of vibration signal will also change, which will result in changing the vibration image. Consequently, performance of CNN model may drop significantly for prediction under different operating conditions. To get the training data from all potential operating conditions may not be feasible for most of the real-world applications. Therefore, there is need to find some signal properties which are invariant to change in operating conditions and only change due to change in health state. So that models trained under one set of operating conditions may predict correctly under different operating conditions. This paper proposes a defect diagnosis method for rolling element bearings, under variable operating conditions (speed and load) based on CNN and order maps. These maps exhibit consistent properties under varying speed; therefore, they can be used to train the CNN model for fault diagnosis under variable speed. Effect of load change on these order maps is experimentally studied and it is found that proposed method can do fault diagnosis of rolling element bearings under variable speed and load with high accuracy.
Spiral bevel gears are known for their smooth operation and high load carrying capability; therefore, they are an important part of many transmission systems that are designed for high speed and high load applications. Due to high contact ratio and complex vibration signal, their fault detection is really challenging even in the case of serious defects. Therefore, spiral bevel gears have rarely been used as benchmarking for gears’ fault diagnosis. In this research study, Artificial Intelligence (AI) techniques have been used for fault detection and fault severity level identification of spiral bevel gears under different operating conditions. Although AI techniques have gained much success in this field, it is mostly assumed that the operating conditions under which the trained AI model is deployed for fault diagnosis are same compared to those under which the AI model was trained. If they differ, the performance of AI model may degrade significantly. In order to overcome this limitation, in this research study, an effort has been made to find few robust features that show minimal change due to changing operating conditions; however, they are fault discriminating. Artificial neural network (ANN) and K-nearest neighbors (KNN) are used as classifiers and both models are trained and tested by using the selected robust features for fault detection and severity assessment of spiral bevel gears under different operating conditions. A performance comparison between both classifiers is also carried out.
Spiral bevel gears are important part of many mechanical transmission systems and are known for their smooth operation and strong load carrying capacity. This type of gear has a high contact ratio, which makes it very difficult to diagnose even serious defects. Therefore, spiral bevel gears have rarely been used as a reference for defect diagnosis techniques. To overcome these challenges, artificial intelligence (AI) techniques are used in this research to diagnose defects in spiral bevel gears. Although Al techniques in the field of fault diagnosis have been very successful, however, these methods largely use the assumption that the training and test data come from the same operating conditions. However, when the operating conditions in which the trained model is deployed for predictions, differ from the operating conditions in which the model was trained, the performance of these approaches might be significantly reduced. Outside the laboratory, in real-world applications, operating conditions significantly vary, and it is difficult to obtain data for all potential operating conditions. To overcome this limitation and to make AI techniques suitable for diagnosing spiral bevel gear faults under different operating conditions, an effort is made to find fault distinguishing features, which are lesser sensitive to operating conditions. Artificial neural network (ANN) and K-nearest neighbors (KNN) are used as classifiers for fault detection. Performance comparison between both classifiers is made to determine their individual capability and suitability for diagnosing defects of spiral bevel gears under different operating conditions.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.