Predictive maintenance, recognized as an effective health management strategy for extending the lifetime of devices, has emerged as a hot research topic in recent years. A general method is to execute two separate steps: data-driven remaining useful life (RUL) prediction and a maintenance strategy. However, among the numerous studies that conducted maintenance and replacement activities based on the results of RUL prediction, little attention has been paid to the impact of preventive maintenance on sensor-based monitoring data, which further affects the RUL for repairable degrading devices. In this paper, an adaptive RUL prediction method is proposed for repairable degrading devices in order to improve the accuracy of prediction results and achieve adaptability to future degradation processes. Firstly, a phased degradation model based on an adaptive Wiener process is established, taking into account the impact of imperfect maintenance. Meanwhile, integrating the impact of maintenance activities on the degradation rate and state, the probability distribution of RUL can be derived based on the concept of first hitting time (FHT). Secondly, a method is proposed for model parameter identification and updating that incorporates the individual variation among devices, integrating maximum likelihood estimation and Bayesian inference. Finally, the effectiveness of the RUL prediction method is ultimately validated through numerical simulation and its application to repairable gyroscope degradation data.

Existing remaining useful life (RUL) prediction methods considering multi-source variability were not applicable to the situation that the uneven measurement interval distribution and inconsistent measurement frequency of degrading equipment. This type of method also has ignored the variability of adaptive drift in the future degradation process. In view of this, based on adaptive Wiener process, the paper proposes a new nonlinear degradation method of the RUL prediction. Firstly, adopting the adaptive Wiener process, we have constructed the nonlinear degradation model with multi-source variability, which randomness of the parameters in the nonlinear function. Secondly, the real-time estimation of multiple hidden states can be realized by the particle filter algorithm. It has derived the RUL distribution in the sense of first hitting time. Using monitoring data of degrading equipment, the adaptive update of model parameters was implemented by expectation maximization algorithm. Finally, the effectiveness and superiority of the proposed model are validated through numerical simulation and lithium-ion battery experiments. The results show that it can effectively improve the prediction accuracy, which has potential application value.

Remote sensing, as a significant technique for acquiring the condition monitoring (CM) information for remaining useful life (RUL) prediction of equipment, primarily consists of two typical steps: sensor monitoring and analog-to-digital conversion (ADC). These two steps may inevitably introduce errors owing to measurement noise and quantization. Unfortunately, existing RUL prediction methods focus on the measurement errors from sensor monitoring but ignore quantization errors from ADC. In this study, we present a nonlinear degradation model that incorporates mixed random noise to estimate the RUL of the stochastic degrading equipment based on observed degradation data involving quantization errors. Through a diffusion process with a nonlinear drift coefficient, the distribution of RUL prediction considering measurement errors and quantization errors is derived. Subsequently, we employ the particle filter to estimate the degradation state based on the observed degradation data, which is affected by quantization error, measurement error, and strong nonlinearity. Additionally, the particle-expectation-maximization (PEM) algorithm is utilized to identify and update model parameters. Finally, numerical simulations verify the effectiveness of our proposed model, and experimental verification is conducted using degradation data from control moment gyroscopes (CMGs). In both cases, the R-square value of the proposed method is no less than 0.98, indicating superior prediction performance compared to the linear method.

The accuracy of lifetime prediction is significantly influenced by the quality of condition monitoring (CM) data, and both forward and reversed problems related to this issue have been extensively investigated thus far. The quality of CM data is inevitably affected by quantization errors resulting from analog-to-digital conversion in practical scenarios; however, it has received relatively less attention compared to measurement error (ME). To clarify the usability of the CM data for predicting the lifetime, a reversed problem — “the specification of the both quantization and ME range to achieve a desirable lifetime prediction performance” is considered in this paper. To address this problem, guidelines for formulating specification limits related to distribution-based quantization and ME characteristics are developed. By referring to a widely employed degradation model based on the Wiener process, permissible values for both ME and quantization parameters under a specified lifetime prediction requirement are derived. It means that the desirable performance for lifetime prediction cannot be achieved if these permissible values are not met. Additionally, several special scenarios are further considered, and the analytical results of permissible values are obtained. Finally, numerical examples and case study of control moment gyroscopes are provided to illustrate the implementation procedure and usefulness of theoretical results.

With the advancement of artificial intelligence, transportation devices driven by autonomous driving technology can efficiently process large-scale data and predict the lifetime through deep learning, thereby enhancing the intelligent operation and maintenance capabilities of such devices. Considering both safety and economic factors associated with multi-sensor-monitored transportation devices, the truncated lifetime test is typically conducted to acquire degradation data for evaluating the device’s lifetime characteristics. In this context, historical monitoring data are typically truncated, posing practical challenges for existing methods, including the absence of lifetime labels and uncertainty in remaining useful life (RUL) prediction. To address this issue, for transportation devices under multi-sensor monitoring, this paper proposes a novel interactive prognosis method that accounts for the influence of truncated data. In the offline training phase, a stacked denoising attention autoencoder is employed to extract health indicators (HIs) from multi-sensor-monitored data in an unsupervised manner, enabling nonlinear degradation modeling based on these indicators. Furthermore, the optimal solution of the deep learning network is obtained by constructing a feedback optimization function linking truncated HIs with a stochastic degradation model. In the online testing phase, the interactive prognosis model trained offline is employed to extract HIs from on-site testing data, and the probability distribution of the RUL is derived based on the nonlinear Wiener process model characterizing the HIs. Meanwhile, Bayesian posterior distribution can be utilized to enable online updating of parameter estimates. Finally, the case studies on turbofan engines are provided to demonstrate the proposed method’s effectiveness.