Automated surface inspection (ASI) is critical to quality control in industrial manufacturing processes. Recent advances in deep learning have produced new ASI methods that automatically learn high-level features from training samples while being robust to changes and capable of detecting different types of surfaces and defects. However, they usually rely heavily on manpower to collect and label training samples. In this paper, a generic semi-supervised deep learning-based approach for ASI that requires a small quantity of labeled training data is proposed. While the approach follows the MixMatch rules to conduct sophisticated data augmentation, we introduce a new loss function calculation method and propose a new convolutional neural network based on a residual structure to achieve accurate defect detection. An experiment on two public datasets (DAGM and NEU) and one industrial dataset (CCL) is carried out. For public datasets, the experimental results are compared against several best benchmarks in the literature. For the industrial dataset, the results are compared against deep learning methods based on benchmark neural networks. The proposed method achieves the best performance in all comparisons. In addition, a comparative experiment of model performance given a different number of labeled samples is conducted, demonstrating that the proposed method can achieve good performance with few labeled training samples.

Catastrophic interlocking chemical accidents (CICAs) are the most frequent hazardous events that occur in chemical parks, and they result in severe human casualties and property losses. Emergency rescue facilities play an important role in ensuring the safety of chemical parks, as their locations determine the efficiency and effectiveness of the emergency management process. This paper proposes a multi-objective emergency rescue facilities location (ERFsL) model for emergency management of CICAs. Due to the special characteristics of CICAs, the proposed model considers not only the spatial characteristics of the facilities but also the potential risks associated with CICAs. A quantitative area risk assessment (QARA) approach based on domino effect and grid partition is involved to delineate the potential negative impacts of CICAs on different demand points. Moreover, a matrix encoding scheme is coupled with Pareto-based multi-objective evolutionary algorithms (MOEAs) for model solving. A case study is presented to demonstrate the methodology. A set of reasonable and non-dominated solutions indicate that the matrix encoding scheme and the corresponding evolutionary operators can be successfully coupled with the MOEAs to produce feasible child solutions. We also find that the involvement of risk analysis has a significant effect on the location decision. The findings show that our work has the potential to ensure the safety and long-term development of the parks.

Heterogeneous vehicle routing problem with time windows (HVRPTW) aspect of hazardous material (HazMat) transportation is studied in this article. Given the multiobjective nature of HazMat transportation, a three-objective optimization model is defined for HVRPTW in HazMat transportation. The objective is to determine fleet size and routes so as to meet all given constraints as well as to minimize the objectives of the total traveling cost, the transportation risk, and the average vehicle redundancy. A load-variant HazMat transportation risk assessment model considering vehicle type and waiting time is presented to describe the transportation risk. A variable neighborhood search-based hybrid multiobjective evolutionary algorithm (VN-HMOEA) is proposed for solving the problem. The proposed VN-HMOEA integrates a two-phase push forward insertion heuristic (TP-PFIH) for initial population construction, specialized evolutionary operators for optimizing different objectives and a VNS metaheuristic for local search exploitation. The algorithm is tested on the modified Solomon benchmark instances for HVRPTW. Experimental results show that the proposed VN-HMOEA is competitive in terms of convergence and diversity. We also find that multiobjective fashion is of great significance for transportation risk mitigation to provide a set of nondominated solution rather than a single solution.

The use of soft sensors is an alternative solution to physical sensors for real-time acquisition of key quality variables that cannot be measured online in real-time. LSTM-based methods have been improving the state-of-the-art in soft sensors. In order to mine the correlation between historical information and current input, and utilize the relatively important historical information, we propose an underlying recurrent neural network unit called information filtering unit (IFU) and an IFU-based LSTM soft sensor (IF-LSTM). In IFU, a feed forward neural network is firstly constructed to predict the hidden state of the previous time step based on the current input and subsequently a scaling factor is obtained, and the hidden state vector is then scaled to highlight those important components that contain non-redundant information with the current input. Finally, an IF-LSTM model is constructed by combining IFU and LSTM units for predicting hard-to-obtain key quality variables from easily-measured process variables, and its performance is validated on two benchmark industrial datasets (SRU and DC datasets). Experimental results show that the performance of the IF-LSTM outperforms popular RNN variants such as LSTM, GRU, Bi-LSTM and Bi-GRU in all metrics (RMSE, MAE, MAPE and R ² ), and achieves an R ² of 0.9929 on the DC dataset, indicating that the predicted values are well-fitted to the true values. Furthermore, comparative experiments with state-of-the-art methods in the literature show that the IF-LSTM model achieves the best RMSE of 0.0069 on the DC dataset, and the second best RMSE of 0.0124 on the SRU dataset.

Nonintrusive load monitoring is a technology that can identify the users’ internal energy consumption by using the data measured at a single point on the bus and event detection is a key technical problem that needs to be solved. An algorithm combining probability and expert heuristic models is proposed for event detection in this study, including an event predetection subalgorithm called voting improved isolated forest (VIIF) for high-sensitivity event predetection and an event verification subalgorithm called time shift downsampling matching (TSDM) for high-accuracy event verification. VIIF is used to detect suspicious events quickly from the low-frequency characteristics of the signal; TSDM identifies real events from suspicious events by analyzing high-frequency characteristics of the signal. To evaluate the proposed algorithm, three datasets are used. Compared with the state-of-the-art algorithms, the proposed algorithm has great adaptability and accuracy to long-transient events and small-signal events.