Prognostics and health management (PHM) increasingly play a constructive role throughout the entire lifetime of industrial equipment, significantly benefiting from extensive research in physical modelling and machine learning techniques. This has led to the development of hybrid approaches that seamlessly integrate both domains through physics-informed machine learning (PIML). PIML ensures the generation of cohesive solutions encompassing various aspects of physics knowledge across different stages of the machine-learning pipeline, substantially contributing to detection, diagnostics, and prognostics. However, PIML’s design relies heavily on expert experience and demands rigorous interdisciplinary expertise, requiring a profound understanding of machine learning and physical principles. Inadequate design of PIML often leads to suboptimal outcomes, where the combined effect is less than the sum of its parts. Currently, PIML lacks a scalable and engineered application architecture to effectively utilise its embedded results. To address this challenge, our paper introduces a novel parallel architectural approach that employs pre-training and fine-tuning strategies for optimising the different model parts. Its data-driven branch is first trained with zero output in the PI branch, then fine-tuning the physics-informed branch. It takes the frozen data-driven model as a fixed feature extractor to get physics-consistency prediction. This approach proposes a generic solution for embedding physics knowledge into ML that guarantees performance improvement. The effectiveness of our approach is validated in the context of Remaining Useful Life (RUL) prediction using MIT-Stanford battery data.

Transformer, built on the self-attention mechanism, has been demonstrated to be effective in numerous applications. However, in the context of prognostics and health management, the self-attention mechanism in the Transformer is not effective in selecting the most important features that are highly correlated with the remaining useful life (RUL) of a component. To address this issue, we developed a novel conditional variational transformer architecture consisting of four networks: two generative networks and two predictive networks. The first generative network uses the transformer encoder–decoder as well as both condition monitoring data and RUL as input to extract the most important features in one feature space from condition monitoring data. The second generative network uses the transformer encoder and condition monitoring data to extract features in another feature space. The two predictive networks use the extracted features in two different feature spaces to make predictions. A KL-divergence is used to minimize the distance between the two feature spaces learned by the first and second generative networks so that the feature space extracted from the second generative network can approximate the feature space extracted from the first generative network. We demonstrated that the proposed method is effective in predicting the RUL of bearings using two datasets.

Graph convolutional networks have been increasingly used to estimate the state of health and predict the remaining useful life of batteries. However, there are two issues with conventional graph convolutional networks. Firstly, they ignore the correlation between features and the state of health or remaining useful life. Secondly, they do not consider temporal relationships among features when projecting aggregated temporal features into another dimensional space. To address these issues, two types of undirected graphs are introduced to simultaneously consider the correlation among features and the correlation between features and the state of health or remaining useful life. A conditional graph convolution network is built to handle these graphs, incorporating a dual spectral graph convolutional operation to analyze the topological structures of these graphs. Additionally, the dilated convolutional operation is integrated with the proposed conditional graph convolution network to account for the temporal correlation among the aggregated features. Two battery datasets were used to evaluate the effectiveness of the presented method, resulting in a minimum mean absolute remaining useful life prediction error of 3.219. Moreover, the proposed method outperforms methods reported in the literature, such as Gaussian processes and other deep learning methods.

Graph Convolutional Networks (GCNs) have recently been used to predict the remaining useful life (RUL) of bearings due to its effectiveness in revealing correlations in condition monitoring data. However, traditional GCNs use a single graph only, either a temporal-correlated graph or a feature-correlated graph without considering both temporal and feature correlations of condition monitoring data. Additionally, traditional GCNs rely heavily on pre-defined graphs to aggregate correlated features. However, the topology of these pre-defined graphs may vary depending on a pre-defined threshold for cosine similarity or covariance which might affect prediction accuracy and robustness. To address these issues, we introduce a spectral graph convolutional operation that can handle both temporal-correlated and feature-correlated graphs, which allows one to consider both the temporal and feature correlations simultaneously. Moreover, we introduce a self-attention mechanism to construct the temporal-correlated and feature-correlated graphs automatically without defining a threshold. Such a mechanism allows the predictive model to learn graphs automatically during training so that the prediction accuracy and robustness can be significantly improved. The proposed method is demonstrated on two bearing datasets, and the experimental results have shown that it outperforms both traditional GCNs and other deep-learning methods in predicting RUL of bearings.

Data fusion integrates data from multiple sources to improve prediction performance. While significant research has been conducted to develop data-level and feature-level fusion methods, very few studies are performed to develop more effective decision-level data fusion methods. This research aims at developing a decision-level data fusion approach that transforms low-dimensional decisions (i.e., predictions) made based on individual sensor data such as temperature and vibration to high-dimensional decisions. Integration of these high-dimensional decisions is formulated as a convex optimization problem rather than a traditional multivariate linear regression problem. The proposed decision-level data fusion approach is demonstrated in two cases: 1) quality control in additive manufacturing and 2) predictive maintenance in aircraft engines. Experimental results have shown that the proposed decision-level fusion method can reduce prediction variance by at least 30% as well as increase prediction accuracy by 45%.

Information security policy (ISP) plays an important role in information security management in organizations. Past research investigated various factors that may impact employee behavior toward security policy compliance from the perspective of general deterrence theory (GDT), protection and motivation Theory (PMT), and rational choice theory (RCT). However, there is no unifying foundation/framework that examines all of those factors in a harmonic way so that the research findings can guide information security practices and research into the employee ISP compliance management context. Additionally, prior findings provided mixed results. This study proposes a research model based on the awareness-motivation-capability (AMC) framework, aiming to unify the factors to predict employee ISP compliance intention. We believe that a harmonic approach in managing employee ISP compliance can create optimal outcomes.

The proliferation of real-time monitoring systems and the advent of Industrial Internet of Things (IIoT) over the past few years necessitates the development of scalable and parallel algorithms that help predict mechanical failures and remaining useful life of a manufacturing system or system components. Classical model-based prognostics require an in-depth physical understanding of the system of interest and oftentimes assume certain stochastic or random processes. To overcome the limitations of model-based methods, data-driven methods such as machine learning have been increasingly applied to prognostics and health management (PHM). While machine learning algorithms are able to build accurate predictive models, large volumes of training data are required. Consequently, machine learning techniques are not computationally efficient for data-driven PHM. The objective of this research is to create a novel approach for machinery prognostics using a cloud-based parallel machine learning algorithm. Specifically, one of the most popular machine learning algorithms (i.e., random forest) is applied to predict tool wear in dry milling operations. In addition, a parallel random forest algorithm is developed using the MapReduce framework and then implemented on the Amazon Elastic Compute Cloud. Experimental results have shown that the random forest algorithm can generate very accurate predictions. Moreover, significant speedup can be achieved by implementing the parallel random forest algorithm.