On-line machinery health diagnosis and prognosis for predictive maintenance and quality assurance of equipment functioning

Machinery health diagnosis and prognosis based on process models and/or process parameters is an important function in automated manufacturing set-ups and safety-critical systems. The primary goal of Intelligent Diagnosis and Prognosis System (IDPS) is continuously and accurately monitoring the current health of critical components in complex equipment, diagnosing their degradation and defect severity, and prognosticating the remaining useful life of the equipment.; This research proposes a theoretical framework for machinery health diagnosis and prognosis. Vibration sensor data is collected and used to indirectly model the equipment. The performance of sensor-based approach is influenced by the parsimonious, yet informative, representation of the sensor signals. We developed efficient signal representation schemes using wavelets, both continuous and discrete, in order to capture the dynamics underlying vibration signals. Then, the signals are visualized in time-frequency domain such that deviant signal structures caused by defective components are clearly characterized.; Based on the features extracted from wavelet analysis, a hierarchical neural networks scheme is devised for diagnosis. This scheme implements sensor-based data fusion and global decision fusion, leading to accurate diagnosis performance for recognizing the defective patterns.; In addition, a novel precursory failure index (PFI) is developed in this work. It provides a way of measuring the significance of equipment failure, detecting the initiation of a fault and extracting impulsive disturbances in vibration signals that reflect equipment anomalies.; For the prediction, an adaptive forecasting procedure based on the time-ordered sequence of PFI is proposed to estimate the future status of equipment. The procedure provides a way for the forecasting model to adapt itself to the underlying process change more accurately and quickly than conventional models and the use of tracking signal makes effective early fault initiation detection possible.; The developed methodologies are described with reference to two gear systems—aft transmission of a helicopter and stand-alone industrial gearbox. This work is the first step towards building a versatile and general IDPS applicable to a wide range of machinery, helping to achieve improved product quality, better plans for maintenance, and quality assurance of equipment functioning.