Stephan W. Wegerich

A fault-tolerant sensory diagnostic system for intelligent vehicle application

A properly designed automotive sensor monitoring and diagnostic system must be capable of detecting and distinguishing sensor and component malfunctions in the presence of signal noise, varying vehicle operating conditions and multiple faults. The technique presented in this paper addresses these objectives through the implementation of a multivariate state estimation algorithm based upon pattern recognition methodology coupled with a statistically-based hypothesis test. Utilizing a residual signal vector generated from the difference between the estimated and measured current states of a system, disturbances are detected and identified with statistical hypothesis testing. Since the hypothesis testing utilizes the inherent noise on the signals to obtain a conclusion and the state estimation algorithm requires only a majority of the sensors to be functioning to ascertain the current state, this technique has proven to be quite robust and fault-tolerant. Several examples of its application are presented.

HIGH PERFORMANCE CONDITION MONITORING OF AIRCRAFT ENGINES

A similarity-based modeling (SBM) technique is demonstrated that provides very early annunciation of the onset of gas path faults in aircraft engines. This powerful approach is shown to provide high fidelity estimates for real-time condition monitoring of aircraft engine signals. These estimates are used to detect the onset of changes in the inter-relationship between the various signals using a sophisticated set of built-in algorithms and tools. The ability of the SBM software to reliably detect subtle changes in signal behavior that are characteristic of a developing anomaly is coupled with a diagnostic rules engine to enable a rapid and robust fault recognition capability. The SBM software operates using a set of algorithms that construct a multivariate nonparametric model of the traditional monitoring sensors (pressure transducers, thermocouples, flow meters, etc.) present in the system. This model is used to generate real-time estimates of sensor values that represent normal system operation. A series of sophisticated tools compares these very high fidelity estimates to the actual sensor readings to detect discrepancies. Finally, a series of logic rules derived from a combination of engineering analysis and experience is applied to the output from the modeling engine in real-time to alert the user of developing serious conditions that need either immediate or planned maintenance attention. The software system provides a complete approach to asset monitoring that minimizes down time, maximizes availability, encodes (preserves) operator knowledge and lowers the overall costs associated with maintaining the assets. In this paper, we demonstrate the use of the similarity-based modeling approach for detecting faults in the gas path of aircraft engines. Some results from the monitoring of over 1,100 engines at a major commercial airline over a two-year period are described. Operationally, the early detection of developing engine faults has prevented delays and cancellations, and has contributed to a reduction in the airline’s in-flight shutdown rate. Financially, this approach has led to significant cost savings by the prevention of major secondary damage.Copyright

Feasibility of personalized nonparametric analytics for predictive monitoring of heart failure patients using continuous mobile telemetry

Nonparametric model-based analytics personalized to the physiology of each patient are investigated for predictive monitoring of exacerbation in heart failure patients at home. Multivariate vital sign data are provided by means of continuous bio-signal acquisition with a mobile phone-based wearable sensor system worn by patients for several hours a day in the home ambulatory environment. Perturbation analysis demonstrates that individual patient physiological behavior is indeed effectively learned by the analytics, with high sensitivity to changes in physiological dynamics. Comparison of the analytics results with absence of unplanned medical events and self-reported wellness during regular patient follow-up demonstrate a very low false alert burden, suggesting this approach is efficient for remote clinical surveillance.