Monitoring of complex industrial processes can be achieved by obtaining process data by utilising various sensing modalities. The recent emergence of deep learning provides a new routine for processing multi-sensor information. However, the learning ability of shallow neural networks is insufficient, and the data amount required by deep networks is often too large for industrial scenarios. This paper provides a novel deep transfer learning method as a possible solution that offers an advantage of better learning ability of the deep network without the requirement for a large amount of training data. This paper presents how Transformer with self-attention trained from natural language can be transferred to the sensor fusion task. Our proposed method is tested on 3 datasets: condition monitoring of a hydraulic system, bearing, and gearbox dataset. The results show that the Transformer trained from natural language can effectively reduce the required data amount for using deep learning in industrial sensor fusion with high prediction accuracy. The difficult and uncertain artificial feature engineering which requires a large workload can also be eliminated, as the deep networks are able to extract features automatically. In addition, the self-attention mechanism of Transformer aids in the identification of critical sensors, hence the interpretability of deep learning in industrial sensor fusion can be improved.

Connected autonomous vehicles (CAVs) offer significant enhancements in coordinated traffic and safety through real-time vehicle-to-vehicle or vehicle-to-infrastructure communications, establishing them as a potent tool for augmenting driving tasks. However, the extensive information-sharing framework inherent in CAVs amplifies the risk associated with sensor anomalies, posing challenges to the reliability and security of the system. Responding to this timely research challenge, this study proposes a novel anomaly detection method, namely Dual-channel Self-attention-based Convolutional Neural Network (DSA-CNN) for multivariate time series data. Through the introduction of the Dual-channel Self-attention Mechanism, DSA-CNN can progressively and autonomously extract spatiotemporal features from multivariate time series data. The proposed method was tested under a variety of common threatening sensor anomaly patterns of CAVs summarised in the literature, and evaluated under multiple different performance metrics. The results demonstrate its advantages in detecting minor anomalies and enhancing sensitivity, outperforming previously reported methods in the literature. Across all experimental scenarios, an average sensitivity improvement of 2.53% was observed, complemented by an average F1 score increase of 1.47%. In CAV settings, maintaining high sensitivity to ensure fewer undetected anomalies, alongside the ability to detect small anomalies, can be more important for the robustness and safety measures of CAV systems.

A green revolution has accelerated over the recent decades with a look to replace existing transportation power solutions through the adoption of greener electrical alternatives. In parallel the digitisation of manufacturing has enabled progress in the tracking and traceability of processes and improvements in fault detection and classification. This paper explores electrical machine manufacture and the challenges faced in identifying failures modes during this life cycle through the demonstration of state-of-the-art machine vision methods for the classification of electrical coil winding defects. We demonstrate how recent generative adversarial networks can be used to augment training of these models to further improve their accuracy for this challenging task. Our approach utilises pre-processing and dimensionality reduction to boost performance of the model from a standard convolutional neural network (CNN) leading to a significant increase in accuracy.

Within aerospace and automotive manufacturing, the majority of quality assurance is through inspection or tests at various steps during manufacturing and assembly. Such tests do not tend to capture or make use of process data for in-process inspection and certification at the point of manufacture. Inspection of the product during manufacturing can potentially detect defects, thus allowing consistent product quality and reducing scrappage. However, a review of the literature has revealed a lack of any significant research in the area of inspection during the manufacturing of terminations. This work utilises infrared thermal imaging and machine learning techniques for inspection of the enamel removal process on Litz wire, typically used for aerospace and automotive applications. Infrared thermal imaging was utilised to inspect bundles of Litz wire containing those with and without enamel. The temperature profiles of the wires with or without enamel were recorded and then machine learning techniques were utilised for automated inspection of enamel removal. The feasibility of various classifier models for identifying the remaining enamel on a set of enamelled copper wires was evaluated. A comparison of the performance of classifier models in terms of classification accuracy is presented. The best model for enamel classification accuracy was the Gaussian Mixture Model with expectation maximisation; it achieved a training accuracy of 85% and enamel classification accuracy of 100% with the fastest evaluation time of 1.05 s. The support vector classification model achieved both the training and enamel classification accuracy of more than 82%; however, it suffered the drawback of a higher evaluation time of 134 s.