With the development of the Industrial Internet of Things (IIoT) technology, fault diagnosis has emerged as a critical component of its operational reliability, and machine learning algorithms play a crucial role in fault diagnosis. To achieve better fault diagnosis results, it is necessary to have a sufficient number of fault samples participating in the training of the model. In actual industrial scenarios, it is often difficult to obtain fault samples, and there may even be situations where no fault samples exist. For scenarios without fault samples, accurately identifying the unknown faults of equipment is an issue that requires focused attention. This paper presents a method for the normal-sample-based mechanical equipment unknown fault detection. By leveraging the characteristics of the autoencoder network (AE) in deep learning for feature extraction and sample reconstruction, normal samples are used to train the AE network. Whether the input sample is abnormal is determined via the reconstruction error and a threshold value, achieving the goal of anomaly detection without relying on fault samples. In terms of input data, the frequency domain features of normal samples are used to train the AE network, which improves the training stability of the AE network model, reduces the network parameters, and saves the occupied memory space at the same time. Moreover, this paper further improves the network based on the traditional AE network by incorporating a convolutional neural network (CNN) and a long short-term memory network (LSTM). This enhances the ability of the AE network to extract the spatial and temporal features of the input data, further improving the network’s ability to extract and recognize abnormal features. In the simulation part, through public datasets collected in factories, the advantages and practicality of this method compared with other algorithms in the detection of unknown faults are fully verified.

In order to improve the efficiency of autonomous robot sorting steel ingots, this paper proposes a two‐dimensional weighted equivalent clustering‐based progressive probabilistic hough transform (2D‐WEC‐PPHT) algorithm for the problem of identifying and locating strip ingots in automatic picking strip ingot palletizing. First, the steel ingot image is preprocessed and the PPHT linear detection method is used to extract the edge linear information of the steel ingot. Second, the angle and position information are normalized and weighted, and a two‐dimensional clustering distance calculation method is proposed for two‐dimensional clustering of the extracted line information. Besides, the clustering line clusters with close angles were processed into equivalent lines by mean fitting. Then, the average threshold method was used to segment the ingot between each two fitting lines, and the minimum rectangle was used to box the segmentation part. The centroid of the box‐selected rectangle is the positioning center of the ingot, and the long side angle of the box‐selected rectangle is the deviation angle information of the ingot. Finally, the experimental results show that the number of redundant lines detected by the 2D‐WEC‐PPHT is significantly less than that of traditional methods such as HT and PPHT. The positioning speed of ingot is faster than that of HT and PPHT, and the processing time is reduced from 20 s to 10 s. In the case of large proportion of old ingots with weak reflection, the recognition accuracy reached 93.5%, and the angular and position positioning accuracy were 2.270° and 11.675 mm, respectively. The recognition accuracy of new ingots with strong reflection reached 99.574%, which met the requirements of picking positioning accuracy.

With the development of the Industrial Internet of Things (IIoT) technology, fault diagnosis has emerged as a critical component of its operational reliability, and machine learning algorithms play a crucial role in fault diagnosis. To achieve better fault diagnosis results, it is necessary to have a sufficient number of fault samples participating in the training of the model. In actual industrial scenarios, it is often difficult to obtain fault samples, and there may even be situations where no fault samples exist. For scenarios without fault samples, accurately identifying the unknown faults of equipment is an issue that requires focused attention. This paper presents a method for the normal-sample-based mechanical equipment unknown fault detection. By leveraging the characteristics of the auto-encoder network (AE) in deep learning for feature extraction and sample reconstruction, normal samples are used to train the AE network. Whether the input sample is abnormal is determined via the reconstruction error and a threshold value, achieving the goal of anomaly detection without relying on fault samples. In terms of input data, the frequency domain features of normal samples are used to train the AE network, which improves the training stability of the AE network model, reduces the network parameters, and saves the occupied memory space at the same time. Moreover, this paper further improves the network based on the traditional AE network by incorporating a convolutional neural network (CNN) and a long short-term memory network (LSTM). This enhances the ability of the AE network to extract the spatial and temporal features of the input data, further improving the network’s ability to extract and recognize abnormal features. In the simulation part, through public datasets collected in factories, the advantages and practicality of this method compared with other algorithms in the detection of unknown faults are fully verified.

This paper proposes an adaptive threshold segmentation algorithm for the magnesium ingot stack based on image overexposure area correction (ATSIOAC), which solves the problem of mirror reflection on the surface of magnesium alloy ingots caused by external ambient light and auxiliary light sources. Firstly, considering the brightness and chromaticity information of the mapped image, we divide the exposure probability threshold into weak exposure and strong exposure regions. Secondly, the saturation difference between the magnesium ingot region and the background region is used to obtain a mask for the magnesium ingot region to eliminate interference from the image background. Then, the RGB average of adjacent pixels in the overexposed area is used as a reference to correct the colors of the strongly exposed and weakly exposed areas, respectively. Furthermore, in order to smoothly fuse the two corrected images, pixel weighted average (WA) is applied. Finally, the magnesium ingot sorting experimental device was constructed and the corrected top surface image of the ingot pile was segmented through ATSIOAC. The experimental results show that the overexposed area detection and correction algorithm proposed in this paper can effectively correct the color information in the overexposed area, and when segmenting ingot images, complete segmentation results of the top surface of the ingot pile can be obtained, effectively improving the accuracy of magnesium alloy ingot segmentation. The segmentation algorithm achieves a segmentation accuracy of 94.38%.