Data-driven Prognostic Approach Research Articles

Remaining useful life prognostics during boiling-induced surface degradation

After sustaining long-term nucleate boiling, any surface is prone to changes over time, which signifies an important operational characteristic for reliable thermal management; however, recent research of this topic is extremely scarce. In this work, we first present long-term pool boiling experiments on samples of different surface topography and morphology in de-ionized water and in an aqueous salt solution. We show the changes induced by vigorous bubble nucleation over the course of several hundred hours of operation, discussing the underlying degradation phenomena. Following this, we present an adaptive data-driven prognostic approach, capable of estimating the remaining useful life (RUL) of a boiling surface in real time, based on change of surface temperature. The method consists of (i) a Kalman filter to identify the degradation drift model and (ii) a Monte Carlo simulation to propagate the drift to the terminal threshold, obtaining a distribution of predicted RUL values. Its practical applicability, validated on experimental results, is reinforced by its advantageous features of (i) prognostic performance not depending on the boiling surface or its degradation mechanisms, (ii) low computational demand and (iii) numerous options for individual tuning and extensions.

Data-driven bearing health management using a novel multi-scale fused feature and gated recurrent unit

Remaining useful life (RUL) prediction plays a crucial role in bearing health management which can guarantee the rotating machinery systems’ safety and reliability. This paper promotes the capability of bearing health management by proposing a novel data-driven prognostic approach. The proposed methodology reveals the underlying deterioration progression by constructing a new health indicator (HI) that fuses distribution-similarity-based multi-scale features. More specifically, a graph named WDgram, to represent the distribution similarity between the analyzed signal and healthy reference at different frequency scales, is developed first using the 1/3-binary tree structure and Wasserstein distance (WD). Then the high-dimensional multi-scale features are fused by the multiobjective grasshopper optimization algorithm (MOGOA) to optimize the HI's monotonicity, robustness, and tenability. By taking advantage of the WDgram and MOGOA, the novel HI is able to thoroughly reveal the underlying degradation characteristics and alleviate random fluctuations which frequently decrease the accuracy of RUL estimation. Subsequently, a gated recurrent unit network, incorporated with the grasshopper optimization algorithm (GOA) to adaptively determine some hyperparameters, is established with high predictive accuracy and generalizability. Experimental validation is conducted to prove the effectiveness of the proposed method. High accuracy and generalizability can be obtained by the proposed method, demonstrating its substantial advantages in health management for rotating machinery systems.

Data-driven prognostics based on time-frequency analysis and symbolic recurrent neural network for fuel cells under dynamic load

Data-centric prognostics is beneficial to improve the reliability and safety of proton exchange membrane fuel cell (PEMFC). For the prognostics of PEMFC operating under dynamic load, the challenges come from extracting degradation features, improving prediction accuracy, expanding the prognostics horizon, and reducing computational cost. To address these issues, this work proposes a data-driven PEMFC prognostics approach, in which Hilbert-Huang transform is used to extract health indicator in dynamic operating conditions and symbolic-based gated recurrent unit model is used to enhance the accuracy of life prediction. Comparing with other state-of-the-art methods, the proposed data-driven prognostics approach provides a competitive prognostics horizon with lower computational cost. The prognostics performance shows consistency and generalizability under different failure threshold settings.

Data-driven Prognostic Approaches for Semiconductor Manufacturing Process: A Review of Recent Works and Future Perspectives

The manufacturing process of semiconductor devices is one of the most complex processes in manufacturing industry. The devices fabrication is performed through hundreds of sequential process steps with different recipes. The level of complexity is also increasing due to the high demands in terms of feature size and number of devices. Maintaining high yield and good quality production are the main objectives of these industries. These objectives can be achieved by adopting efficient maintenance strategies. In this context, a suitable prognostic model is required in order to schedule the maintenance actions. Among the different prognostic approaches, data-driven ones received a lot of attention since they do not require any specific knowledge for modeling these complex processes. Although the advances in data-driven prognostic works, there is a real lack of survey papers that overview and discuss the existing approaches for this industry over the past 10 years. Therefore, this paper presents a systematic overview of data-driven prognostic for semiconductor manufacturing. It investigates the different used methods, the challenges of their application and the unexplored research areas.

A comparative Study of Data-driven Prognostic Approaches under Training Data Deficiency

In this study, a remaining useful life (RUL) prognostic performance is explored for the three algorithms under the insufficient training datasets: data augmentation prognostics (DAPROG) developed by the authors, and trajectory similarity-based prediction (TSBP) and multilayer perceptron (MLP) based neural network, which is one of the popular approaches in the literature. The Commercial Modular Aero-Propulsion System Simulation (CMAPSS) datasets are taken to assess each algorithm. Only a fraction is chosen randomly from the training dataset to simulate the data deficiency and evaluate the prognostic performance. As a result, the DAPROG is found to be the most superior in terms of the prediction error when the data deficiency is severe. The prediction variance for the chosen training dataset is also small, indicating more uniform performance.

Data-driven failure prediction and RUL estimation of mechanical components using accumulative artificial neural networks

It is necessary to develop advanced fault prognostics techniques for improving maintenance planning and scheduling to avoid unexpected shutdowns, additional expenses and decreased productivity. Such techniques include advanced failure prediction and remaining useful life (RUL) estimation of mechanical components comprising manufacturing systems with complicated structures. This study proposes a novel data-driven prognostic analysis approach for predicting the failure of a mechanical component based on its degradation path and estimating the RUL. A simulated labelled degradation dataset of a mechanical component with a predefined failure threshold was exploited. In order to increase the ability to maintain the increasing trend and the monotony of the degradation path, supervised machine learning models, including combined artificial neural network architectures and an improved version of the neuron-by-neuron training algorithm using accumulative neural networks design were applied for the prediction process.The expected degradation path was extrapolated as a testing dataset of the trained prediction model using an accumulative function. The predicted values are updated with each new data point during the training process until a failure occurs. The results show that the used approach is efficient in predicting the failure and estimating the RUL of mechanical components with high accuracy and a high prediction success rate, and it can maintain the monotonous trend of the degradation path. On the other hand, the used network architectures enable the prediction of the failures of mechanical components within a manufacturing system having a complex structure and providing a vast amount of data.

RUL prediction of machinery using convolutional-vector fusion network through multi-feature dynamic weighting

Based on the features extracted from the condition monitoring data, data-driven prognostic approaches are able to predict the remaining useful life (RUL) of machinery. Existing methods usually assume that a certain feature contributes consistently to the prediction results during the operation. In fact, the degradation sensitivity of each feature varies with time in most practical cases, which limits the prediction accuracy of RUL. To tackle this issue, a novel convolutional-vector fusion network (C-VFN) is proposed in this paper. A vector-dynamic weighted fusion (V-DWF) algorithm is designed to dynamically evaluate the degradation sensitivity of each feature over time. The fluctuations of feature sensitivities over time are visualized through a weight map. Then, the sensitivity weights are assigned to the corresponding features to estimate the RUL. Meanwhile, the insensitive features are iteratively eliminated through a mechanism of RUL-result-oriented feedback. The proposed model is validated using accelerated degradation data of axle reducers and XJTU-SY datasets. The experimental results show that the C-VFN is able to estimate the degradation sensitivity of each feature along with time and improve the accuracy of RUL prediction.

Relation between prognostics predictor evaluation metrics and local interpretability SHAP values

Maintenance decisions in domains such as aeronautics are becoming increasingly dependent on being able to predict the failure of components and systems. When data-driven techniques are used for this prognostic task, they often face headwinds due to their perceived lack of interpretability. To address this issue, this paper examines how features used in a data-driven prognostic approach correlate with established metrics of monotonicity, trendability, and prognosability. In particular, we use the SHAP model (SHapley Additive exPlanations) from the field of eXplainable Artificial Intelligence (XAI) to analyze the outcome of three increasingly complex algorithms: Linear Regression, Multi-Layer Perceptron, and Echo State Network. Our goal is to test the hypothesis that the prognostics metrics correlate with the SHAP model's explanations, i.e., the SHAP values. We use baseline data from a standard data set that contains several hundred run-to-failure trajectories for jet engines. The results indicate that SHAP values track very closely with these metrics with differences observed between the models that support the assertion that model complexity is a significant factor to consider when explainability is a consideration in prognostics.

Data-driven prognostic model for temperature field in additive manufacturing based on the high-fidelity thermal-fluid flow simulation

The process-structure–property relationship for additive manufacturing (AM) is typically derived starting from the temperature profile which can be achieved by the meso-scale thermal-fluid flow simulation with huge computational cost. We propose a data-driven prognostic approach with specialized physical constraints to rapidly predict the temperature profiles. The dataset is constructed from the physics-based thermal-fluid flow simulation results under different manufacturing parameters. The temperature field around the molten pool region is statistically characterized by the function parameters of the individual isotherms, which are essentially the output of the data-driven model based on the input manufacturing parameters, while the temperature field is reconstructed using the interpolation approach based on the predicted isotherms. The data-driven predicted temperature profiles are validated against those from the thermal-fluid flow simulations, and then further applied in the thermal stress and grain growth simulations, of which the results are compared with those using the temperature profile directly from thermal-fluid flow simulations. The results demonstrate that our data-driven approach is highly feasible in predicting the geometry features of the isotherms and temperature profiles around the molten pool regions.

Remaining useful life prediction of PEMFC based on cycle reservoir with jump model

Accurate prognosis of limited durability is one of the key factors for the commercialization of proton exchange membrane fuel cell (PEMFC) on a large scale. Thanks to ignoring the structure of the PEMFC and simplifying the prognostic process, the data-driven prognostic approaches was the commonly used for predicting remaining useful life (RUL) at present. In this paper, the proposed cycle reservoir with jump (CRJ) model improves the ESN model, changes the connection mode of neurons in the reservoir and speeds up the linear fitting process. The experiment will verify the performance of CRJ model to predict stacks voltage under static current and quasi-dynamic current conditions. In addition, the reliability of the CRJ model is verified with different amount of data as the training and test sets. The experimental results demonstrate that the CRJ model can achieve better effect in the remaining useful life prognosis of fuel cells.

Recent advances and trends of predictive maintenance from data-driven machine prognostics perspective

In the Engineering discipline, prognostics play an essential role in improvingsystemsafety, reliability and enabling predictive maintenance decision-making. Due to the adoption of emerging sensing techniques and big data analytics tools, data-driven prognostic approaches are gaining popularity. This paper aims to deliver an extensive review of recent advances and trends of data-driven machine prognostics, with a focus on their applications in practice. The primary purpose of this review is to categorize existing literature and report the latest research progress and directions to support researchers and practitioners in acquiring a clear comprehension of the subject area. This paper first summarizes fundamental methodologies on data-driven approaches for predictive maintenance. Then, the article further conducts a comprehensive investigation on the different fields of applications of machine prognostics. Finally, a discussion on the challenges, opportunities, and future trends of predictive maintenance is presented to conclude this paper.

Sensor-Data-Driven Fusion Prognostic Framework for Complex Engineered Systems

Complex engineered systems have complex system structures and competing failure mechanisms, which means that neither model-based or data-driven approaches are suitable for conducting health state prognostics. Prognostics and health management (PHM) has been found to be an effective tool for advance failure warnings, system health status assessments, and predictive maintenance. To improve system health state forecasting accuracy, this paper developed a PHM-based fusion prognostic framework that strategically integrated the strengths of a model-based particle filter approach and a data-driven prognostics approach and obviated the respective limitations. An entropy-based fusion prognostics model was employed to fuse the particle filter prognostics results generated at different times and obtain more accurate prognostics results. Using sensor data sets from the NASA Ames Research Center, the developed fusion prognostics framework was then employed to estimate the health state of an aircraft gas turbine engine. The experimental results demonstrated that the proposed fusion prognostics framework was an effective prognostics tool and could achieve more accurate, robust health state prognostics.

A Hybrid Prognostic Approach Based on Deep Learning for the Degradation Prediction of Machinery

Remaining useful life (RUL) prediction is of great significance for prognostic and health management (PHM) as it can achieve more reliable and effective maintenance strategies. With the advances in the field of deep learning, data-driven methods have provided promising prognostic prediction results. Hence, this research presents a data-driven prognostic approach based on deep learning models for predicting the RUL of mechanical systems effectively. Multiple separable convolution layers, a bidirectional Long Short-Term Memory (LSTM) layer, and fully-connected layers (FCL) are included in the proposed network, named the SC-BLSTM, to accomplish more accurate prognostic prediction from the raw degradation data acquired by different sensors. The proposed SC-BLSTM approach aims to learn complex and nonlinear features from the input data and capture temporal dependencies from the learned features. The presented approach in this research is tested and verified on the degradation data of turbofan engines (C-MAPSS dataset) from NASA. The result demonstrated that the SC-BLSTM is able to achieve more effective RUL prediction compared with some existing prognostic models.

A hybrid prognostic method based on gated recurrent unit network and an adaptive Wiener process model considering measurement errors

Remaining useful life (RUL) prediction is fundamental to prognostics and health management (PHM). Considering the advantages of both model-based and data-driven prognostic approaches, this paper develops a hybrid prognostic method for machinery degradation. First, a 3σ criterion-based algorithm is introduced to detect the initial timepoint of degradation. Second, gated recurrent unit (GRU) network is utilized to learn the degradation characteristics based on the available data and thereby predict the long-term degradation trend by a multi-prediction procedure. Then, an adaptive Wiener process model considering measurement errors is constructed. The states of this model consisting of the drift rate and the underlying degradation value are updated adaptively based on the monitored observations and the predictions by GRU using a Kalman filtering algorithm. The predicted values of the RUL can be determined according to the underlying degradation and the failure threshold. Finally, to account for the drift adaptivity in the future degradation, exponentially weighted average method is adopted to aggregate the estimated drift sequence from the current time until failure for the derivation of real-time RUL distributions. The effectiveness and superiority are illustrated by a simulation study and an application to rolling element bearings.

Remaining useful life prediction of PEMFC systems under dynamic operating conditions

The Prognostic and Health Management (PHM) has been developed for more than two decades. It is capable to anticipate the impending failures and make decisions in advance to extend the lifespan of the target systems, such as Proton Exchange Membrane Fuel Cell (PEMFC) systems. Prognostic is a critical stage of PHM. Among various prognostic methods, the data-driven ones could predict the system lifespan based on the device’s knowledge and historical data. In the Remaining Useful Life (RUL) prediction, the Health Indicators (HIs) should be able to reflect the health states of the PEMFC stack. Moreover, an effective HI could help to define an explicit degradation state and improve the prediction accuracy. The HIs of voltage and power are usually used under static conditions due to their monotonic decreasing characteristics. Besides, the measurements of voltage and current are implemented easily in practice. Nevertheless, the static HIs are unable to be directly used under the dynamic operating conditions because they are sensitive to the mission profiles. To overcome the weakness of static HIs, a convenient and practical HI named Relative Power-loss Rate (RPLR) is proposed herein. According to the polarization curve at the beginning of life, the initial power under different mission profiles can be identified. Then the actual power is obtained by monitoring the current and voltage continuously. Finally, the RPLR is calculated based on the initial power and actual power. Afterward, the RUL of PEMFC is predicted by some Artificial Intelligence (AI) prognostic algorithms. Among the various data-driven prognostic approaches, Echo State Network (ESN) has provided an efficient and promising solution for the RUL prediction of PEMFC systems. Compared with classical Recurrent Neural Network (RNN), it could accelerate the convergence rate and reduce the computational complexity. Nevertheless, the traditionally used single-input ESN structure is feeble to handle the varying mission profiles. As a scheduling variable, the current is an interesting parameter since it represents the working properties to some extent. Considering the system’s dynamic characteristics, the stack current is regarded as another input of ESN, and the output matrix’s dimension is increased at the same time. Therefore, a double-input ESN structure is proposed to enhance the prediction performance. Based on the dynamic HI of RPLR, three dynamic micro-cogeneration (μ-CHP) durability tests of PEMFC systems are used to verify the improved ESN prediction structure.

Intelligent prognostics of machining tools based on adaptive variational mode decomposition and deep learning method with attention mechanism

In the modern manufacturing industry, remaining useful life (RUL) prediction of the machining tools plays a significant role in promoting machining efficiency, ensuring product quality and reducing production costs. In recent years, many data-driven prognostic approaches have been developed for machining tools, but few of them have considered the operating conditions such as spindle load and rotating speed that may have great impact on the degradation behavior and sensor signals. It may give rise to more uncertainty and lead to an obvious decrease in prediction accuracy when operating condition changes. Besides, feature extraction from the raw signals that are nonstationary, nonlinear, and mixed with abundant noise is essential but quite challenging. To address these issues, this paper proposes a novel prognostic approach for machining tools under dynamic operating condition with varying spindle load. In the proposed approach, an adaptive variational mode decomposition (VMD) is newly developed to adaptively search the optimal parameters for processing the raw vibration data, then several components with good trendability and noise robustness are obtained for feature extraction. Furthermore, a deep learning model combining one-dimensional convolutional long short-term memory (LSTM) with attention mechanism is constructed to perform RUL prediction. Numerical experiments on a real-world case study show the effectiveness and superiority of the proposed approach in comparison with other baseline data-driven approaches.

An ensemble prognostic method for lithium-ion battery capacity estimation based on time-varying weight allocation

Capacity estimation is of great significance to help assess the performance degradation of lithium-ion batteries, so as to take actions to extend their lifetime. Traditional capacity estimation methods for Lithium-ion batteries are usually based on individual model-based or data-driven prognostic approaches. However, no single prognostic method performs appropriately for all possible situations as each individual method presents particular assumptions and application limitations. Therefore, this paper presents an ensemble prognostic framework that combines multiple individual prognostic algorithms to improve the accuracy and robustness of battery capacity estimation. In the proposed ensemble prognostic framework, the degraded capacity data of the full battery life cycles are divided into three parts: a training dataset, a validation dataset, and a test dataset, among which the training and validation datasets are employed for member prognostic model training, the validation dataset is utilized for weight calculation, and the test dataset is used for prognostic performance assessment. A validation-data based induced ordered weighted averaging (IOWA) operator, i.e. V-IOWA operator, is proposed to realize time-varying weight assignment. By summing the weighted prognostic results of each member prognostic algorithm, the ensemble prognostic results are finally obtained. Effectiveness of the proposed approach was validated based on datasets provided by NASA Ames Prognostics Center of Excellence. The experiment results indicated that the proposed ensemble prognostic approach outperforms individual prognostic algorithms with a higher accuracy.

Adaptive Machine Learning Approach for Fault Prognostics based on Normal Conditions - Application to Shaft Bearings of Wind Turbine

Prognostics can enhance the reliability and availability of industrial systems while reducing unscheduled faults and maintenance cost. In real industrial systems, data collected from the normal operation conditions of system is available, but there is a lack of historical degradation data is often unavailable. Hence, this paper proposes a general data-driven prognostic approach dealing with the lack of degradation data in the offline phase. First, features are computed on the collected raw signal, then One Class Support Vector Machine (OCSVM) is used to detect the degradation, this anomaly detection method is trained using only normal operation data. Then, features are ranked according to the selection criteria. The feature having the highest score is chosen as Health Indicator (HI). Finally an adaptive degradation model is applied for the prediction of the degradation evolution over time and Remaining Useful Life (RUL) estimation. The proposed approach is validated using run-to-failure vibration data collected from a high speed shaft bearings of a commercial wind turbine.

Deep separable convolutional network for remaining useful life prediction of machinery

Abstract Deep learning is gaining attention in data-driven remaining useful life (RUL) prediction of machinery because of its powerful representation learning ability. With the help of deep learning techniques, the machine degradation information can be mined more sufficiently and some promising RUL prediction results have been achieved in several studies recently. These deep learning-based prognostics approaches, however, have the following weaknesses: 1) Their prediction performance largely depends on the hand-crafted feature design. 2) The correlations of different sensor data are not explicitly considered in representation learning. To overcome the above weaknesses, a new deep prognostics network named deep separable convolutional network (DSCN) is proposed in this paper for RUL prediction of machinery. In the proposed DSCN, the monitoring data acquired by different sensors are directly used as the inputs of the prognostics network. Then, a separable convolutional building block with a residual connection is built based on separable convolutions and squeeze and excitation operations. Through stacking multiple separable convolutional building blocks, the high-level representations are automatically learned from the input data. Finally, the RUL is estimated by feeding the learned representations into the fully-connected output layer. The proposed DSCN is validated using the vibration data from accelerated degradation tests of rolling element bearings and the public degradation simulation data of turbine engines, respectively. The experimental results show that the proposed DSCN is able to provide accurate RUL prediction results based on the raw multi-sensor data and is superior to some existing data-driven prognostics approaches.

Valve Failure Prognostics in Reciprocating Compressors Utilizing Temperature Measurements, PCA-Based Data Fusion, and Probabilistic Algorithms

In this paper, temperature measurements are utilized to develop health indicators based on principal component analysis toward the probabilistic estimation of the remaining useful life (RUL) of reciprocating compressors in service. Temperature degradation histories obtained from 13 actual valve failure cases constitute the training data in a data-driven prognostic approach. Two data-driven prognostic methodologies are presented and proposed based on probabilistic mathematical models, i.e., gradient boosted trees and nonhomogeneous hidden semi-Markov models. The training and testing process of all models is described in detail. RUL prognostics in unseen data are obtained for all models. Beyond the mean estimates of the RUL, the uncertainty associated with the point prediction is quantified and upper/lower confidence bounds are also estimated. Prediction estimates for 12 real-life failure cases are presented and the pros and cons of each model's performance are highlighted. Several metrics are utilized to assess the performance of the prognostic algorithms and conclusions are drawn regarding the prognostic capabilities of each of them.