An Integrated Prognostics Method Under Time-Varying Operating Conditions

Abstract

In this paper, we develop an integrated prognostics method considering a time-varying operating condition, which integrates physical gear models and sensor data. By taking advantage of stress analysis in finite element modeling (FEM), the degradation process governed by Paris' law can adjust itself immediately to respond to the changes of the operating condition. The capability to directly relate the load to the damage propagation is a key advantage of the proposed integrated prognostics approach over the existing data-driven methods for dealing with time-varying operating conditions. In the proposed method, uncertainties in material parameters are considered as sources responsible for randomness in the predicted failure life. The joint distribution of material parameters is updated as sensor data become available. The updated distribution better characterizes the material parameters, and reduces the uncertainty in life prediction for the specific individual unit under condition monitoring. The update process is realized via Bayesian inference. To reduce the computational effort, a polynomial chaos expansion (PCE) collocation method is applied in computing the likelihood function in the Bayesian inference and the predicted failure time distribution. Examples based on crack propagation in a spur gear tooth are given to demonstrate the effectiveness of the proposed method. In addition, the example also shows that the proposed approach is effective even when the current loading profile is different from the loading profile under which historical data were collected.Index Terms-Bayesian update, integrated prognostics, polynomial chaos expansion, time-varying operating condition, uncertainty quantification. ACRONYMS AND ABBREVIATIONS: FEM finite element modeling PCE polynomial chaos expansion RUL remaining useful life PHM proportional hazard model FE finite element SIF stress intensity factor Manuscript

Data-Driven Battery Lifetime Prediction and Confidence Estimation for Heavy-Duty Trucks

“…A key difference in the data compared to many other prognostic approaches, e.g., [2,5,8], is that only one snapshot per vehicle is available and it is not possible to track the vehicle to predict failure time. The lifetime function (1) is proposed as an estimate of the battery lifetime and the RSF model output is the estimate of the reliability function which can be used to compute the lifetime function estimate (2).…”

“…There are two main differences between IJ variance estimate of the RF model compared to variance estimate of lifetime function (8). First, the output of the RF model is either a class or regression value, but in the RSF case the output is a time dependent function, and secondly, the lifetime function is a ratio of the reliability estimatesR V (t) as in (2).…”

“…In turn, nonparametric data-driven models use machine learning methods that do not have any basic assumption regarding underlying degradation distribution [7]. Nowadays, hybrid methods that fuse predictions both from the model-based and data-driven approaches are proposed, see [8,9]. Unlike the aforementioned methods, where in most cases time series of sensor data is available, the data set under study only contains information retrieved from a vehicle during one of its workshop visits.…”

Integrated Equipment Health Prognosis Considering Crack Initiation Time Uncertainty and Random Shock

“…Recently, integrated prognostics methods [17][18][19][20][21][22][23][24][25][26] were developed to achieve real-time RUL prediction during the system operations by combining both condition monitoring (CM) data and physics of failure. The integrated methods usually have an updating process by assimilating observations, during which the uncertainty is expected to shrink so that the confidence increases in the predicted results.…”

“…Prognostic algorithms have been proposed in a large number of publications for various industrial applications. Those algorithms are mostly either data-driven [3][4][5][6][7], physically motivated [8][9][10][11][12][13][14][15][16] or model and data integrated [17][18][19][20][21][22][23][24][25][26].…”