Scene reconstruction has attracted widespread attention due to its extensive applications in intelligent devices. Recently, 3D Gaussian Splatting (3DGS) has gained recognition as a prominent technique owing to its impressive pixel-level rendering quality and high reconstruction speed. However, 3DGS scenes constructed with only photometric supervision are highly prone to RGB overfitting. This problem neglects the depth of objects and the surface normal in the reconstructed scene, which leads to the lack of geometric consistency. To address these issues, this study introduces a coupled optimization of depth and normal within the 3D Gaussian Splatting (CODN-GS) framework. Firstly, a normal–depth–normal transformation is applied to ensure accurate capture of geometric features in the reconstructed scenes. Secondly, a robust monocular depth supervision model generates depth maps that are refined via global and local adjustments, which serve as guidance for the model to accurately learn scene geometry. Thirdly, a normal supervision model is incorporated as a complement to depth supervision, jointly optimizing the overall scene geometry. Finally, comprehensive experiments on the Replica, MipNerf360, and ScanNet datasets demonstrate that CODN-GS reduces RMSE-D and RMSE-N by at least 9% and 13%. These results confirm that the proposed method outperforms state-of-the-art methods in both depth and normal accuracy.

3-D reconstruction of transparent objects is a pivotal technology extending the application domain in intelligent robotics. However, during the reconstruction process, neural radiance field (NeRF) based on deep learning tends to blur the boundaries of transparent objects and overlook highlights, leading to a decrease in reconstruction performance. To overcome this issue, this article investigates a new NeRF for transparent and highlighted objects (NeRF-THO). First, a multiresolution transparent feature aggregated module is established to process sparse RGB images. Second, a trainable separation-reconstruction module is designed to decompose the scene for modeling by two independent NeRFs. Finally, a carefully designed truncated signed distance function generation network is devised to achieve reasonable separation results and expedite the sampling process. A realistic synthetic dataset of transparent glass bottles is established to evaluate the model’s performance. Experimental results demonstrate that NeRF-THO exhibits superior performance in the quality of 3-D reconstruction.

Image anomaly detection and segmentation are important for the development of automatic product quality inspection in intelligent manufacturing. Because the normal data can be collected easily and abnormal ones are rarely existent, unsupervised methods based on reconstruction and embedding have been mainly studied for anomaly detection. But the detection performance and computing time require to be further improved. This article proposes a novel framework, named as pretrained feature mapping (PFM), for unsupervised image anomaly detection and segmentation. The proposed PFM maps the image from a pretrained feature space to another one to detect the anomalies effectively. The bidirectional and multihierarchical bidirectional PFM are further proposed and studied for improving the performance. The proposed framework achieves the better results on well-known MVTec AD dataset compared with state-of-the-art methods, with the area under the receiver operating characteristic curve of 97.5% for anomaly detection and of 97.3% for anomaly segmentation over all 15 categories. The proposed framework is also superior in terms of the computing time. The extensive experiments on ablation studies are also conducted to show the effectiveness and efficiency of the proposed framework.

Surface defect detection (SDD) is critical in the smart manufacturing systems to ensure product quality. Nevertheless, the defective samples are always insufficient, and there exists high-noise backgrounds in the samples from the real-world SDD applications, which could significantly affect SDD models. To remedy the drawbacks, a new foreground-perception cycle-consistent adversarial network (FCGAN) is proposed, and it can synthesize high-quality pseudodefect images by recognizing the foreground of samples. First, the attention networks are combined with the generator and discriminator to guide the training process. Second, the discriminative foreground is recognized and used to synthesize coordinated and realistic pseudodefect images. Third, the adversarial training for FCGAN is designed. The experiments are carried out on two well-known datasets and a real-world SDD for glass bottles. The results show that FCGAN can produce better pseudodefect images than the state-of-the-art generative adversarial networks, and it can promote the detection accuracy of SDD as well.

Surface defect detection is the essential process to ensure the quality of products. Surface defect classification (SDC) based on deep learning (DL) has shown its great potential. However, the well-trained SDC model usually requires large training data, and the small intraclass differences between the defect and normal samples also degrades the performance of SDC model. To overcome these drawbacks, this article proposed a new cycle-consistent adversarial networks with attention mechanism (AttenCGAN). First, AttenCGAN is used for synthesizing defect samples to enlarge the samples volume. Second, the attention mechanism is adopted for the feature enhancement by finding the discriminative parts of the samples and enlarging the differences among the samples. AttenCGAN is tested on KolektorSDD and DAGM2007 datasets, and its accuracies are 98.53% and 99.57% with only a few samples. The experiment results show that AttenCGAN outperforms other published SDC methods based on DL and machine learning, which validates its potential.

Deep learning (DL) has been widely applied in the fault diagnosis field. However, the depth of DL models in fault diagnosis is very shallow compared with benchmark convolutional neural network (CNN) models for ImageNet. But it is hard to train a very deep CNN model without the large amount well-organized datasets like ImageNet. In this research, a new transfer learning based on pre-trained VGG-19 (TranVGG-19) is proposed for fault diagnosis. Firstly, a time-domain signals to RGB images conversion method is proposed. Then, the pre-trained VGG-19 is applied as feature extractor to obtained the features of converted images. Finally, a softmax classifier is trained on the features. The proposed TranVGG-19is tested on the famous motor bearing dataset from Case Western Reserve University. The final prediction accuracy of TCNN is 99.175% and the training time of TranVGG-19is only near 200 seconds. These results outperform many DL and machining learning methods.

Fault diagnosis plays an important role in modern industry. With the development of smart manufacturing, the data-driven fault diagnosis becomes hot. However, traditional methods have two shortcomings: 1) their performances depend on the good design of handcrafted features of data, but it is difficult to predesign these features and 2) they work well under a general assumption: the training data and testing data should be drawn from the same distribution, but this assumption fails in many engineering applications. Since deep learning (DL) can extract the hierarchical representation features of raw data, and transfer learning provides a good way to perform a learning task on the different but related distribution datasets, deep transfer learning (DTL) has been developed for fault diagnosis. In this paper, a new DTL method is proposed. It uses a three-layer sparse auto-encoder to extract the features of raw data, and applies the maximum mean discrepancy term to minimizing the discrepancy penalty between the features from training data and testing data. The proposed DTL is tested on the famous motor bearing dataset from the Case Western Reserve University. The results show a good improvement, and DTL achieves higher prediction accuracies on most experiments than DL. The prediction accuracy of DTL, which is as high as 99.82%, is better than the results of other algorithms, including deep belief network, sparse filter, artificial neural network, support vector machine and some other traditional methods. What is more, two additional analytical experiments are conducted. The results show that a good unlabeled third dataset may be helpful to DTL, and a good linear relationship between the final prediction accuracies and their standard deviations have been observed.

Welding, an irreplaceable process in the modern manufacturing industry, consumes enormous amounts of energy. The schedule in a welding shop greatly impacts both its energy consumption and productivity. Thus, it is of great significance to solve the welding shop scheduling problem (WSSP) considering both energy efficiency and productivity. In this paper, to solve a real-life WSSP, a multiobjective mathematical model is proposed and an effective multiobjective artificial bee colony algorithm (MOABC) is developed. The results of a designed numerical experiment indicate that the proposed MOABC performs better than Strength Pareto Evolutionary Algorithm 2 and Nondominated Sorting Genetic Algorithm II. Finally, the proposed model and MOABC algorithm are applied to solve a real-life girder WSSP of a Chinese crane company. The results also demonstrate that the proposed method can greatly reduce energy consumption and makespan compared to other algorithms.

Surface roughness has a great influence on the product properties. Predicting the surface roughness is an important work for modern manufacturing industry. In this paper, a novel prediction method called Free Pattern Search (FPS) is proposed to explicitly construct the surface roughness prediction model. FPS takes the advantage of the expression tree in gene expression programming (GEP) to encode the solution and to expresses a non-determinative tree using a fixed length individual. FPS is inspired by Pattern Search (PS) and hybrid a scatter manipulator to keep the diversity of the population. Three machining parameters, the spindle speed, feed rate and the depth of cut are used as the independent input variables when prediction the surface roughness in end milling. Experiments are conducted to verify the performance of FPS and FPS obtains good results compared with other algorithm. The predictive model found by FPS agrees with the experimental result. The variable relations are also showed in the predictive model, and the results shows that they are fit to the experiments well.

Cutting forces is one of the most fundamental elements that affect the performance of cutting operation. Finding the rules that how process and environment factors affect the values of cutting forces will help to set the process parameters of the future cutting operation and further improve production quality and efficiency. Since cutting forces is impacted by different machining parameters and the inherent uncertainties in the machining process, how to predict the cutting forces becomes a challengeable problem for the researchers and engineers. Gene Expression Programming (GEP) combines the advantages of the genetic algorithm (GA) and genetic programming (GP), and has been successfully applied in function mining and formula finding, so it should be suitable to solve the above problem. In this paper, a method based on GEP has been proposed to construct the prediction model of cutting forces in a face-milling operation. At the basis of defining a GEP environment for the problem and improving the method of constant creation, an explicit prediction model of cutting forces has been constructed. To verify the feasibility and performance of the proposed approach, experimental studies have been conducted to compare this approach with some previous works. The obtained results show that the constructed prediction model fits very well with the experimental data, and can be used to estimate the cutting forces and optimize the cutting parameters. The proposed method will lead to the reduction in production costs and production time, and improvement of product quality.