个人简介
研究兴趣
Three-dimensional subsurface defect shape reconstruction and visualisation by pulsed thermography
Defects detected by most thermographic inspection are represented in the form of 2D image, which might limit the understanding of where the defects initiate and how they grow over time. This paper introduces a novel technique to rapidly estimate the defect depth and thickness simultaneously based on one single-side inspection. For the first time, defects are reconstructed and visualised in the form of a 3D image using cost-effective and rapid pulsed thermography technology. The feasibility and effectiveness of the proposed solution is demonstrated through inspecting a composite specimen and a steel specimen with semi-closed airgaps. For the composite specimen, this technique can deliver comparatively low averaged percentage error of the estimated total 3D defect volume of less than 10
The implication of non-driving activities on situation awareness and take-over performance in level 3 automation
The driver’s take-over performance is of great importance for driving safety in conditionally automated driving since the driver is required to respond appropriately to control the vehicle if there is a system failure. The engagement of different non-driving activities (NDAs), considered as the main factor of the driver’s take-over performance has been investigated in this study from both perspectives of the driver’s situation awareness and take-over quality. The activities are divided into 2 groups, which are active interaction mode and passive interaction mode based on the engagement of human and object. The results suggest that the engagement of NDAs could reduce the driver’s situation awareness. Driver’s attention level is different for each activity. Particularly, active interaction mode NDAs requests more mentally demanding and drivers are not sensitive to the driving situation change when they are doing such activities. In addition, there is no significant difference in the maximum lateral error with NDAs engagement. However, it takes more time to achieve a safe control transition for drivers who are doing the NDAs. The active interaction mode NDAs request even more time. Moreover, the transition process could benefit from steering wheel haptic feedback torque, which can be considered as an effective take-over assistance system.
Recurrent neural networks and its variants in Remaining Useful Life prediction
Data-driven techniques, especially artificial intelligence (AI) based deep learning (DL) techniques, have attracted more and more attention in the manufacturing sector because of the rapid growth of the industrial Internet of Things (IoT) and Big Data. Tremendous researches of DL techniques have been applied in machine health monitoring, but still very limited works focus on the application of DL on the Remaining Useful Life (RUL) prediction. Precise RUL prediction can significantly improve the reliability and operational safety of industrial components or systems, avoid fatal breakdown and reduce the maintenance costs. This paper reviews and compares the state-of-the-art DL approaches for RUL prediction focusing on Recurrent Neural Networks (RNN) and its variants. It has been observed from the results for a publicly available dataset that Long Short-Term Memory (LSTM) networks and Gated Recurrent Unit (GRU) networks outperform the basic RNNs, and the number of the network layers affects the performance of the prediction.
Inspection of electronic component using pulsed thermography
Counterfeit electronic components (CEC) are of great concern to governments and industry globally as they could lead to systems and mission failure, malfunctioning of safety critical systems, and reduced reliability of high-hazard assets. Depending on the cost of CEC going into the production line, some industries might look to have some sort of inspection capability in-house to screen critical components before they go to assembly. Although advanced counterfeit inspection methods have been developed for a variety of components, they generally exhibit a combination of disadvantages such as destructive, low throughput, high unit cost, or inaccessible to unskilled operator. This paper investigates the potential of pulsed thermography on detection of CEC in a fast and non-destructive manner. The second derivative of post-heat thermal response is used to construct a fingerprint to differentiate genuine and counterfeit components. Results successfully demonstrate the potential of the proposed solution.
Detectability evaluation of attributes anomaly for electronic components using pulsed thermography
Counterfeit Electronic Components (CECs) pose a serious threat to all intellectual properties and bring fatal failure to the key industrial systems. This paper initiates the exploration of the prospect of CEC detection using pulsed thermography (PT) by proposing a detectability evaluation method for material and structural anomalies in CECs. Firstly, a numerical Finite Element Modelling (FEM) simulation approach of CEC detection using PT was established to predict the thermal response of electronic components under the heat excitation. Then, by experimental validation, FEM simulates multiple models with attribute deviations in mould compound conductivity, mould compound volumetric heat capacity and die size respectively considering experimental noise. Secondly, based on principal components analysis (PCA), the gradients of the 1st and 2nd principal components are extracted and identified as two promising classification features of distinguishing the deviation models. Thirdly, a supervised machine learning-based method was applied to classify the features to identify the range of detectability. By defining the 90 classification accuracy as the detectable threshold, the detectability ranges of deviation in three attributes have been quantitively evaluated respectively. The promising results suggest that PT can act as a concise, operable and cost-efficient tool for CECs screening which has the potential to be embedded in the initial large scale screening stage for anti-counterfeit.
Automated tortuosity analysis of nerve fibers in corneal confocal microscopy
Precise characterization and analysis of corneal nerve fiber tortuosity are of great importance in facilitating examination and diagnosis of many eye-related diseases. In this paper we propose a fully automated method for image-level tortuosity estimation, comprising image enhancement, exponential curvature estimation, and tortuosity level classification. The image enhancement component is based on an extended Retinex model, which not only corrects imbalanced illumination and improves image contrast in an image, but also models noise explicitly to aid removal of imaging noise. Afterwards, we take advantage of exponential curvature estimation in the 3D space of positions and orientations to directly measure curvature based on the enhanced images, rather than relying on the explicit segmentation and skeletonization steps in a conventional pipeline usually with accumulated pre-processing errors. The proposed method has been applied over two corneal nerve microscopy datasets for the estimation of a tortuosity level for each image. The experimental results show that it performs better than several selected state-of-the-art methods. Furthermore, we have performed manual gradings at tortuosity level of four hundred and three corneal nerve microscopic images, and this dataset has been released for public access to facilitate other researchers in the community in carrying out further research on the same and related topics.
An uncertainty quantification and aggregation framework for system performance assessment in industrial maintenance
The exponential increase in technological complexity of modern engineering systems necessitates rigorous and accurate maintenance planning to determine optimum equipment availability and turnaround time whilst allowing for overruns and unforeseen costs. Quality and availability of quantitative data, as well as qualitative expert opinion and experience expose uncertainties that can result in under or over estimation of the above factors. Uncertainty quantification in complex engineering systems should consider inter-connected components and associated processes from a combination of quantitative and qualitative (compound) perspectives. This paper presents a framework to quantify and aggregate compound uncertainties and to be assessed against a predetermined acceptable level of uncertainty. This will provide maintenance planners with a confident, comprehensive view of parameters surrounding the above factors to improve decision making capabilities. The framework was validated by assessing individual and compound uncertainties in a bespoke heat exchanger test rig comprised of subsystem modules interact in a non-linear manner, as well as subjective opinions and actions of operators. The results demonstrate the framework’s ability to effectively quantify these factors with an indication of their impact on the system. Future work will include further validation with more complex case studies and development of methods to forecast the quantified uncertainty through the in-service phase of an asset’s life cycle.
An adaptive pig face recognition approach using convolutional neural networks
The evolution of agriculture towards intensive farming leads to an increasing demand for animal identification associated with high traceability, driven by the need for quality control and welfare management in agricultural animals. Automatic identification of individual animals is an important step to achieve individualised care in terms of disease detection and control, and improvement of the food quality. For example, as feeding patterns can differ amongst pigs in the same pen, even in homogenous groups, automatic registration shows the most potential when applied to an individual pig. In the EU for instance, this capability is required for certification purposes. Although the RFID technology has been gradually developed and widely applied for this task, chip implanting might still be time-consuming and costly for current practical applications. In this paper, a novel framework composed of computer vision algorithms, machine learning and deep learning techniques is proposed to offer a relatively low-cost and scalable solution of pig recognition. Firstly, pig faces and eyes are detected automatically by two Haar feature-based cascade classifiers and one shallow convolutional neural network to extra high-quality images. Secondly, face recognition is performed by employing a deep convolutional neural network. Additionally, class activation maps generated by grad-CAM and saliency maps are utilised to visually understand how the discriminating parameters have been learned by the neural network. By applying the proposed approach on 10 randomly selected pigs filmed in farm condition, the proposed method demonstrates the superior performance against the state-of-art method with an accuracy of 83 facilitate the real-application of AI-based animal identification in swine production.