| The advancement and adoption of information technology has increasingly influenced consumers' perception on product innovation, quality, variety and diversity, precision and speed of delivery. The increased customer demands enhance the expectations that the industrial manufacturing sector to become agile, and eventually reach abilities for predictive manufacturing. Intensive research and new technologies enable the transformations and technological development in various manufacturing applications, including CNC machine tools, automatic assembly line, semiconductor, material handling, etc. Among the numerous systems or components in a manufacturing equipment, motion control system plays an important role and its accuracy directly determines the precision capability of the whole manufacturing system. Each element in the motion control system requires a self-aware capability. One of the most important components in motion control system is the ball screw; it is the key component to execute the controller command precisely. Thus, this research is mainly focused on the component level -- ball screw health monitoring. This condition-based monitoring study allows the visualization of the ball screw's performance degradation, eventually preventing unexpected failures.;Among many common degradation modes, this thesis focuses on 2 major failure modes - lubrication starvation and preload loss, which happen very frequently and are vital to the ball screw's performance. Improper lubrication can result in increased friction levels so as to generate large amount of heat that can cause ball screw overheating and damage the internal surface or components. The preload is an indicator for the position accuracy loss, but it can also be a factor that can worsen its performance. The continuous operational use can wear out the internal contact surface of ball screw, thus widening the gap between rolling elements and grooves leading to backlash and then precision loss. In this study, laboratory experiments were conducted on full size ball screws to simulate these two types of degradation modes.;The first aspect of this research is focused on comparing the sensor-less method and sensor-rich methods for the early detection of lubrication starvation and evaluation of four health assessment algorithms, Mahalanobis distance (MD), self-organizing map-minimum quantization error (SOM-MQE), logistic regression (LR), and Principal Component Analysis (PCA) -- Hotelling's T 2. Only motor output speed and torque signals were acquired in the sensor-less tests; these four algorithms are evaluated based on how early they can detect the incipient failure. The motor current output (proportional to torque) is a measure of the overall degradation and can be due to many different types of failures; thus, the sensor-rich method is considered to identify the failure modes. The MD, SOM-MQE algorithms are calculated based on the vibration data collected from the accelerometers on the ball screw. This thesis uses the health trend (pattern) to identify the different failure modes. And then it also provides a mechanism for lubrication diagnosis for online lubrication monitoring system.;The second aspect is to study the ball screw performance at different preload levels. By analyzing time domain features and frequency spectrum of vibration signals, this research identifies the preload feature to differentiate backlash case from preload case. Beside from comparing the health value distribution pattern based on SOM-MQE and MD methods, it also applies PCA and SVM method to identify the preload levels, which gives 100% accuracy rate. |
