Machine Prognostics Research Articles

Attention-based ConvNeXt with a parallel multiscale dilated convolution residual module for fault diagnosis of rotating machinery

Convolutional Neural Networks have promoted development of the fault diagnosis in the machine prognostics and health management. However, the existing methods have limited applicability under strong noisy conditions. We propose an attention-based ConvNeXt with a parallel multiscale dilated convolution residual module for the rotor fault diagnosis. The parallel multiscale dilated convolution residual module is firstly introduced to filter noise and extract multiscale discriminative features intelligently. The multi-head attention module is then embedded in ConvNeXt for global discriminative features adaptively. Focal Loss is finally implemented for identifying hard-to-classify samples to further improve the diagnostic accuracy. The application results demonstrate the superiority and robustness of our method in two case studies.

Attention and Long Short-Term Memory Network for Remaining Useful Lifetime Predictions of Turbofan Engine Degradation

Machine Prognostics and Health Management (PHM) is often concerned with the prediction of the Remaining Useful Lifetime (RUL) of assets. Accurate real-time RUL predictions enable equipment health assessment and maintenance planning. In this work, we propose a Long Short-Term Memory (LSTM) network combined with global Attention mechanisms to learn RUL relationships directly from time-series sensor data. We use the NASA Commercial Modular Aero- Propulsion System Simulation (C-MAPPS) datasets to assess the performance of our proposed method. We compare our approach with current state-of-the-art methods on the same datasets and show that our results yield competitive results. Moreover, our method does not require previous degradation knowledge, and attention weights can be used to visualise temporal relationships between inputs and predicted outputs.

Empirically Measuring Transfer Distance for System Design and Operation

Classical machine learning approachesare sensitive to nonstationarity. Transfer learning can address nonstationarity by sharing knowledge from one system to another, however, in areas like machine prognostics and defense, data are fundamentally limited. Therefore, transfer learning algorithms have little, if any, examples from which to learn. Herein, the authors suggest that these constraints on algorithmic learning can be addressed by systems engineering. We formally define transfer distance in general terms and demonstrate its use in empirically quantifying the transferability of models. We consider the use of transfer distance in the design of machine rebuild procedures to allow for transferable prognostic models. We also consider the use of transfer distance in predicting operational performance in computer vision. Practitioners can use the presented methodology to design and operate systems with consideration for the learning theoretic challenges faced by component learning systems.

Machine prognostics under varying operating conditions based on state-space and neural network modeling

Prognostics based on deep learning models usually assume that training and testing data come from the same equipment with similar working conditions. In practice, some new working conditions of test equipment may not be recorded in the training dataset, and thus the learned remaining useful life (RUL) prediction model may not work well. This work is the first to propose a novel prognostic model based on state-space modeling and reinforcement learning to predict the RUL of equipment operating under extrapolated new conditions without corresponding training data. Instead of directly constructing a supervised model that relates monitoring measurements to their RUL, a discrete-time state-space model is built using possible failure histories. The contribution of this paper is twofold. First, on the basis of the state-space model, an interpretable prognostic model that combines Lyapunov constraint and reinforcement learning is proposed to predict the equipment RUL. Second, an adversarial training method based on the idea of H∞ robustness is integrated to reduce the effect of state-space modeling errors. The proposed model aims to reduce the causality between RUL and operating parameters and increase the causality between RUL and unobserved degradation characteristics. Consequently, the proposed model is interpretable and has the potential to predict RUL for the equipment operating under conditions beyond the record. The proposed approach is demonstrated and validated using a simulation study and an experimental dataset.

Spatiotemporal denoising wavelet network for infrared thermography-based machine prognostics integrating ensemble uncertainty

Infrared thermography (IRT) is increasingly deployed in noncontact condition monitoring, and further facilitates the machinery remaining useful life (RUL) prediction. For IRT-based prognostics, the interpretability of degradation features, the spatiotemporal denoising capability, and the effectiveness of uncertainty quantification are crucial issues. Therefore, this paper proposes an ensemble of deep spatiotemporal denoising wavelet networks (EDSDWN). Firstly, a 4D wavelet convolution layer (WCL) is designed to capture crucial degradation-related features with meaningful physical interpretability based on multi-resolution analysis. Secondly, noises existent in features are further filtered out in a restoration process by a proposed deep image stream denoiser (DISD) block with residual learning. A base DSDWN consisting of the WCL, the DISD, and a regressor is comprehensively constructed. Ultimately, the effective uncertainty quantification is implemented by EDSDWN adaptively fitting a RUL density with a Student-t mixture distribution. EDSDWN is constructed by employing an automatically weighted algorithm to incorporate multiple base DSDWNs. Experimental results on a thermal image dataset acquired from rotating machinery and a simulated dataset have proven that EDSDWN is tailored to IRT-based machine prognostics. This proposed method implements considerably lower prediction errors and higher effectiveness of uncertainty quantification than several advanced prognostics models.

A novel dual-stream self-attention neural network for remaining useful life estimation of mechanical systems

Remaining useful life (RUL) estimation plays a crucial role in evaluating health states and improving maintenance plans of mechanical systems. Recently, artificial intelligence-based data-driven methods that use monitoring data as input have made significant progress in machine prognostics. However, current methods commonly ignore the correlations and internal differences of monitoring data, consequently leading to limited estimation performance. Therefore, this paper proposes a novel data-driven RUL estimation method named Dual-Stream Self-Attention Neural Network (DS-SANN). First, the multi-head self-attention mechanism is employed to learn correlations between different monitoring data and weigh the features dynamically to obtain global degraded information. Then, a dual-stream structure network is established to extract features from the original and auxiliary data simultaneously to make a comprehensive reflection of health states. The original and auxiliary data represent absolute values and internal differences of monitoring data, respectively. Finally, the multilayer perceptron is adopted to fuse the obtained features and estimate RUL. In addition, the effectiveness of DS-SANN is validated by the public degradation dataset of turbine engines. Compared with several existing prognostics methods, DS-SANN shows better estimation performance when averaging across all sub-datasets. Specifically, estimation effects evaluated by RMSE and Score improve 21.77% and 32.67%, respectively.

Predictions of component Remaining Useful Lifetime Using Bayesian Neural Network

The Machine Prognostics and Health Management (PHM) are concerned with the prediction of the Remaining Useful Lifetime (RUL) of assets. Accurate real-time RUL predictions are necessary when developing an efficient predictive maintenance (PdM) framework for equipment health assessment. If correctly implemented, a PdM framework can maximize the interval between maintenance operations, minimize the cost and number of unscheduled maintenance operations, and improve overall availability of the large facilities like nuclear power plants (NPPs). This is especially important for nuclear power facilities to maximize capacity factor and reliability. In this work, we propose a data-driven approach to make predictions of both the RUL and its uncertainty using a Bayesian Neural Network (BNN). The BNN utilizes the Bayes by backprop algorithm with variational inference to estimate the posterior distribution for each trainable parameter so that the model output is also a PDF from which one can draw the mean prediction and the associated uncertainty. To learn the correlations between various time-series sensor data measurements, a time window approach is implemented with a two-stage noise filtering process for incoming sensor measurements to enhance the feature extraction and overall model performance. As a proof of concept, the NASA Commercial Modular Aero Propulsion System Simulation (C-MAPPS) datasets are utilized to assess the performance of the BNN model. The modeled system can be treated as a surrogate for turbine generators used in NPPs due to the similar mode of operation, degradation, and measurable variables. Comparisons against other state-of-the-art algorithms on the same datasets indicate that the BNN model can not only make predictions with comparable level of accuracy, but also offer the benefit of estimating uncertainty associated with the prediction. This additional uncertainty, which can be continuously updated as more measurement data are collected, can facilitate the decision-making process with a quantifiable confidence level within a PdM framework. Additional advantages of the BNN are showcased, such as providing component maintenance ranges and model executing frequency, with an example of how the BNN estimated uncertainty can be used to support the continuous predictive maintenance. A PdM framework based on a BNN will allow for utilities to make more informed decisions on the optimal time for maintenance so that the loss of revenue can be minimized from planned and unplanned maintenance outages.

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.

Infrared image stream based regressors for contactless machine prognostics

In a practical production environment, machinery operators would consider using contactless sensing technology to monitor machinery degradation condition due to the concern of interference on production. A stream of time-series infrared images as a contactless sensing technology could capture the spatial and temporal information of a machinery degradation process. However, due to four dimensions (4D) of image stream tensor data, most existing remaining useful life (RUL) prediction methods are not capable of processing this kind of data. To fill this gap, image stream based regressors consisting of a neural network and multiple weighted time window policy are proposed. In the regressors, neural networks are utilized and trained to model the correlation between degradation image stream and its associated RUL. The networks are respectively designed based on some primary neural network blocks, including a fully-connected layer, long short-term memory network (LSTM) and convolutional neural network (CNN). To achieve the processing capability of 4D data, various input preprocessing means of raw image stream data are investigated. Meanwhile, to increase the prediction accuracy of the trained networks, a multiple weighted time window (MWTW) policy is developed. The policy aims to use whole monitoring data rather than a recent time window in existing neural network based RUL prediction methods. The proposed image stream based regressors are validated by using two datasets of degradation infrared images. Results showed that the regressors can predict RUL based on image stream well and MWTW policy has a significant effect on the increase of the prediction accuracy.

Development of Multivariate Failure Threshold with Quantifiable Operation Risks in Machine Prognostics

Prognostics and Health Management (PHM) is attracting the attention from both academia and industry due to its great potential to enhance the resilience and responsiveness of the equipment to the potential operation risks. In literature, many methodologies are proposed to predict the Remaining Useful Life (RUL) of the equipment. However, there are two major challenges that limit the practicality of these methodologies. 1) How to generate a quantifiable Health Indicator (HI) to represent the operation risks? 2) How to define a reasonable failure threshold to predict RUL? To answer these two questions, this paper proposes a novel methodology for failure threshold determination with quantifiable operation risk in machine prognostics. In the proposed methodology, Fisher distance and Mann-Kendall (MK) test are firstly used to extract useful sensors based on which HI is estimated by applying Principle Component Analysis (PCA). Then, Rao-Blackwellized Particle Filter (RBPF) is employed to obtain the HI prediction and the uncertainties. Afterwards, a Bivariate-Weibull-distribution-based risk quantification model is designed to quantify the cumulative risk over time and over the increase of HI. The failure threshold, which is the ending point of the RUL, varies over different users and applications depending on the level of risk they want to tolerate. The validation of the methodology is based on the C-MAPSS data from the PHM data competition 2008 hosted by PHM society. The results validate the effectiveness of the proposed risk quantification method and its potential application on machine prognostics.

Analysis of different RNN autoencoder variants for time series classification and machine prognostics

Recurrent neural network (RNN) based autoencoders, trained in an unsupervised manner, have been widely used to generate fixed-dimensional vector representations or embeddings for varying length multivariate time series. These embeddings have been demonstrated to be useful for time series reconstruction, classification, and creation of health index (HI) curves of machines being used in industrial applications, based on which the remaining useful life (RUL) of machines can be estimated. In this study, we extend the traditional form of RNN autoencoders as a feature extractor for multivariate time series to a more general form in terms of arranging the order of input or output sequences and the hidden unit architecture. We apply the embeddings obtained by different variants of RNN autoencoders for a time series classification task and a machine RUL estimation problem using two publicly available datasets. A random research strategy is used to find the optimal hyperparameters of all variants for each task in order to give a fair comparison of the general performance among different variants over a large hyperparameter space, as well as the best performance that each variant can achieve compared with the best reported values in the literature. Our results show that traditional reversing the order of output time series while maintaining the order of input time series when training an RNN autoencoder does not show improved performance for the two studied cases. Thus, intentionally arranging the input or output order seems unnecessary for training the RNN autoencoder as a feature extractor of time series. We further observe that only the RNN architectures with gating mechanism can achieve the functionality of encoding for the time series, and none of the three common gated architectures we studied shows significantly and consistently improved performance compared to the others on the two studied cases. However, the bidirectional RNN autoencoders yield slightly better performance than their unidirectional counterparts.

Transfer learning for remaining useful life prediction based on consensus self-organizing models

The traditional paradigm for developing machine prognostics usually relies on generalization from data acquired in experiments under controlled conditions prior to deployment of the equipment. Detecting or predicting failures and estimating machine health in this way assumes that future field data will have a very similar distribution to the experiment data. However, many complex machines operate under dynamic environmental conditions and are used in many different ways. This makes collecting comprehensive data very challenging, and the assumption that pre-deployment data and post-deployment data follow very similar distributions is unlikely to hold.In this work, we present a feature-representation based transfer learning (TL) method for predicting Remaining Useful Life (RUL) of equipment, under scenarios that samples with previously unseen conditions are presented in the target domain and the labels are available only for the source domain, but not the target domain. This setting corresponds to generalizing from a limited number of run-to-failure experiments performed prior to deployment into making prognostics with data coming from deployed equipment that is being used under multiple new operating conditions and experiencing previously unseen faults. We employ a deviation detection method, Consensus Self-Organizing Models (COSMO), to create transferable features for building the RUL regression model. These features capture how different a particular equipment is in comparison to its peers.The efficiency of the proposed TL method is demonstrated using the NASA Turbofan Engine Degradation Simulation Data Set. Models using the COSMO transferable features show better performance than other methods on predicting RUL when the target domain is more complex than the source domain.

Data-driven prognosis method using hybrid deep recurrent neural network

Prognostics and health management (PHM) has attracted increasing attention in modern manufacturing systems to achieve accurate predictive maintenance that reduces production downtime and enhances system safety. Remaining useful life (RUL) prediction plays a crucial role in PHM by providing direct evidence for a cost-effective maintenance decision. With the advances in sensing and communication technologies, data-driven approaches have achieved remarkable progress in machine prognostics. This paper develops a novel data-driven approach to precisely estimate the remaining useful life of machines using a hybrid deep recurrent neural network (RNN). The long short-term memory (LSTM) layers and classical neural networks are combined in the deep structure to capture the temporal information from the sequential data. The sequential sensory data from multiple sensors data can be fused and directly used as input of the model. The extraction of handcrafted features that relies heavily on prior knowledge and domain expertise as required by traditional approaches is avoided. The dropout technique and decaying learning rate are adopted in the training process of the hybrid deep RNN structure to increase the learning efficiency. A comprehensive experimental study on a widely used prognosis dataset is carried out to show the outstanding effectiveness and superior performance of the proposed approach in RUL prediction.

A similarity based methodology for machine prognostics by using kernel two sample test

This paper proposes a novel similarity-based algorithm for Remaining Useful Life (RUL) prediction and a methodology for machine prognostics. In the proposed RUL prediction algorithm, a Similarity Matching Procedure including the Kernel Two Sample Test (KTST) is developed to query similar run-to-failure (R2F) profiles from historical data library. Next, the preliminary predictions of RUL are obtained as remaining time-to-failure from the similar R2F records. In the last step, Weibull analysis is performed to fuse the preliminary predictions and to obtain the probability distribution of RUL. Moreover, a methodology for machine prognostics is developed based on the RUL prediction algorithm. Compared with existing similarity-based methods for RUL prediction, the proposed method holds several advantages: 1) the similarities between sensor readings or feature matrices are directly measured without extra health assessment procedure; 2) the proposed method presents good probabilistic interpretations of the prediction uncertainties; 3) the estimated RUL distribution is statistically sound by applying KTST to prescreening the historical R2F records. The effectiveness and the superiority of the proposed method are justified based on the public aero-engine dataset.

Railway Point Machine Prognostics Based on Feature Fusion and Health State Assessment

This paper presents a condition monitoring approach for point machine prognostics to increase the reliability, availability, and safety in railway transportation industry. The proposed approach is composed of three steps: 1) health indicator (HI) construction by data fusion, 2) health state assessment, and 3) failure prognostics. In Step 1, the time-domain features are extracted and evaluated by hybrid and consistency feature evaluation metrics to select the best class of prognostics features. Then, the selected feature class is combined with the adaptive feature fusion algorithm to build a generic point machine HI. In Step 2, health state division is accomplished by time-series segmentation algorithm using the fused HI. Then, fault detection is performed by using a support vector machine classifier. Once the faulty state has been classified (i.e., incipient/starting fault), the single spectral analysis recurrent forecasting is triggered to estimate the component remaining useful life. The proposed methodology is validated on in-field point machine sliding-chair degradation data. The results show that the approach can be effectively used in railway point machine monitoring.

Edge-Based RNN Anomaly Detection Platform in Machine Tools

ABSTRACTWith the rapid advances in machine learning algorithms and sensing technologies, machine prognostics and health management (PHM) via data-driven approaches has become a trend in sophisticated machine tool industry. The run-to-failure data are necessary for data-driven approaches. However, the average life of the machine is two to three years, the time of collecting data is extended. It is a big challenge to collect run-to-failure data and build a PHM model. Therefore, we propose an Edge-based RNN Anomaly Detection Platform (ERADP). ERADP builds the model based on healthy data and notify anomalies two hours in advance. The true alarm rate is up to 100%. Besides, ERADP can accelerate the training time almost 120 times faster than the traditional model.

A Digital Twin Proof of Concept to Support Machine Prognostics with Low Availability of Run-To-Failure Data

Abstract The present research illustrates a Digital Twin Proof of Concept to support machine prognostics with Low Availability of Run-to-Failure Data. Developed in the scope of the Industry 4.0 Lab of the Manufacturing Group of the School of Management of Politecnico di Milano, the Digital Twin is capable to run in parallel to the drilling machine operations and, as such, it enables to predict the evolution of the most critical failure mode, that is the imbalance in the drilling axis. The real-time monitoring of the drilling machine is realized with a low-cost and retrofit solution, which provides the installation of a Raspberry-Pi accelerometer, able to enhance the extant automation. Relying on a joint use of real-time monitoring and simulation, the Digital Twin implements a random coefficient statistical method through the so-called Exponential Degradation Model, eventually demonstrating to increase the prediction precision as monitoring data arrives. The Digital Twin Proof of Concept is described according to the entire process from data acquisition to Remaining Useful Life prediction, following the MIMOSA OSA-CBM standards.

A statistical methodology for the design of condition indicators

Abstract Recent studies in the field of diagnostics and prognostics of machines have highlighted the key role played by non-stationarity – often in the form of cyclostationarity – or non-Gaussianity – often in the form of impulsiveness as characteristic symptoms of abnormality. Traditional diagnostic and prognostic indicators (e.g. the kurtosis) are however sensitive to both types of symptoms without being able to differentiate them. In an effort to investigate how the signal characteristics evolve with the different phases of machine components degradation, this paper proposes a new family of condition indicators able to track cyclostationary or non-Gaussian symptoms independently. A statistical methodology based on the maximum likelihood ratio is introduced as a general framework to design condition indicators. It arrives with the possibility of setting up statistical thresholds, as needed for a reliable diagnosis. The methodology is validated with numerically generated signals and applied to the dataset made available by NSF I/UCR Center for Intelligent Maintenance Systems (IMS). This particular application shows high potential for bearing prognostics by providing condition indicators able to describe different phases of the bearing degradation process.

Machine health management in smart factory: A review

In this paper, we present a review of machine health managements for the smart factory. As the Industry 4.0 leads current factory automation and intelligent machines, the machine health management for diagnostic and prognostic purposes are essential, and their importance is getting more significant for the realization of the smart factory in the Industry 4.0. After brief introductions to important concepts and definitions composing smart factory and Industry 4.0, the developments in maintenance strategies towards Prognostics and health management (PHM) of machines are summarized. The review of machine health managements is followed, classifying the references by the monitoring components, types of measurements, as well as PHM tools and algorithms. 94 existing articles are reviewed and summarized in this regard. The implementations of machine health managements within the smart factory are discussed in terms of data connectivity, communications, Cyber-physical system (CPS) and virtual factory, relating them to Internet of things (IoT), cloud computing, and big data management.

An ontology-based model for prognostics and health management of machines

Abstract Recent advances in smart manufacturing open up opportunities in industrial support, specifically in maintenance and physical asset management. This trend allows data collected from machines in operation to interact with cyberspace computers through a communication network, thus forming the concept of cyber-physical systems (CPS). Besides, rapid advances in information and communications technologies provide approaches for analysing data, in an increasingly rapid, autonomously, ubiquitous and in real time way, providing information that assists humans in making better decisions. In this sense, Prognostics and Health Management (PHM) of machines, is indicated as a promising application of Smart Manufacturing in the CPS context, demanding the standardization of concepts, terms, and a formal implementation of data collection and treatment. For this purpose, the Design Science Research (DSR) methodology is used in this paper, encompassing international standards, the unified 5-level architecture, ontology, and dependability for failure analysis in mechanical components. In addition, the creation of an ontology using the OWL 2 language was guided by the ‘Ontology Development 101′ approach. A pilot test was carried out using a centrifugal pump to demonstrate the applicability of the ontology. Thus, the ontology is evaluated in Protege, which allow queries with SPARQL language to provide future decision-making for condition-based maintenance in real processes.