Instantaneous parameters of rotating machinery contain substantial health state information. Acquiring the instantaneous parameters quickly and accurately is crucial for the fault diagnosis of rotating machinery. Traditional instantaneous parameter detection methods require intervention in the equipment system, which will result in problems such as equipment system structure changes, large instantaneous parameter measurement errors, and inaccurate fault diagnosis results. Hence, a non-contact instantaneous parameter detection method using visual tracking transformation for rotating machinery fault diagnosis is proposed in this paper. First, a sinusoidal modulation pattern is designed and attached to the surface of a rotating shaft as a feature encoder. Subsequently, the state information of the shaft is encoded into an image to obtain an image sequence. Then, the image sequence is decoded using the image tracking transform, and the image sequence information is converted into a series of instantaneous feature information, while the instantaneous parameters are derived by calculation. Finally, an experimental setup for detecting instantaneous parameters of rotating machinery was built. And the instantaneous angle and instantaneous angular speed are experimentally verified. The experimental results show that the proposed method can extract the instantaneous parameters effectively, and the proposed method has the advantages of non-contact measurement and high resolution.

The automation and intelligent scheduling of coal mine auxiliary transport robots, along with high-precision, robust, real-time position sensing technology to ensure inherent safety and production efficiency, are key components of smart mine construction. However, due to the coupling interference of low light and weak textures underground, traditional measurement methods experience reduced accuracy or even failure. Especially in the process of constructing high-precision 3D images of roadways, the accuracy of pose estimation is significantly reduced, which seriously affects the stitching quality and global consistency of 3D point clouds of roadways. In order to solve the above problems, a real-time pose measurement method for roadway robots based on the Retinex principle and phase correlation analysis proposed. In this algorithm, an adaptive discrimination module for low-light images based on image quality evaluation function is designed, which avoids the over-processing of normal images by Retinex. Retinexformer is used as the front-end enhancement module to improve the image clarity and contrast through the deep network, and enhance the perception ability of the algorithm in low-light environment. The image registration is realized by Fourier-Merlin transform (FMT) combined with phase correlation technology, which effectively compensates for the influence of image blur and illumination change caused by sensor motion on pose measurement, and greatly improves the accuracy and reliability of image registration. The real-time measurement of the position and posture of the robot body in the complex roadway environment is realized, which effectively improves the overall robustness and accuracy of the system. Several experiments have been carried out on the EuRoC public dataset and roadway scenes. The results show that compared with the existing measurement algorithms, the RMSE of MH04 and MH05 sequences under low light conditions is reduced by 28.57% and 26.94%, respectively. In order to further verify the generalization of the algorithm, the algorithm is tested in four modes, which shows higher estimation accuracy, real-time and robustness, and improves the pose measurement effect of the sensor in the low-light area of the roadway. Provides theoretical guidance and practical support for the realization of accurate, robust and lightweight autonomous positioning and roadway space reconstruction of coal mine auxiliary transportation robots.

Artificial intelligence techniques play a pivotal role in intelligent fault diagnosis of rotating machinery. Vibration signals from such machinery exhibit strong noise and pronounced time-varying characteristics due to intense interference, which limits the ability of conventional convolutional neural networks to effectively extract features from these noisy, non-stationary time-series signals. To address this challenge, a novel method is proposed that integrates a mode spectral array map (MSAM) with a dual-channel adaptive scaled convolutional neural network for fault diagnosis. First, multiresolution and spectral analyses are applied to the acquired time-series signals to obtain spectral modal component signals. These components are then transformed into images and array-ordered using the Gramian angular summation field (GASF) to construct the MSAM-GASF sample set. Second, a dual-branch convolutional neural network architecture is developed to learn distinct MSAM weight values, enabling high-dimensional feature complementation and enhancement. Finally, multi-fault experiments are conducted on rotating machinery. Experimental results demonstrate that the proposed method achieves high fault recognition accuracy and exhibits superior stability and robustness under strong noise and interference conditions. Consequently, the method shows strong potential for effectively performing intelligent fault diagnosis of rotating machinery.

To address the issue of low diagnostic model accuracy and generalization caused by state data distribution differences and sample imbalance across operating conditions and equipment, this paper proposes a cross-domain imbalanced sample fault diagnosis method based on parallel Temporal Convolutional Attention Network (TCAN) and Transformer, by integrating feature enhancement and transfer learning strategies. First, Refined Composite Hierarchical Fuzzy Dispersion Entropy (RCHFDE) is employed to extract features from state signals in both source and target domains, constructing an RCHFDE feature space to enhance state signals at the data level. Subsequently, a parallel TCAN-Transformer diagnostic model is designed and constructed, trained using complete, normal RCHFDE feature spaces from a specific operating condition or device. The trained parallel TCAN-Transformer diagnostic model is then transferred to new operating conditions or equipment. Using the RCHFDE feature space from limited unbalanced data of the new conditions or equipment, the diagnostic model undergoes parameter fine-tuning and training to obtain the final TL-TCAN-Transformer model suitable for the new context. This transferred diagnostic model is derived from the TCAN-Transformer model through parameter adjustment. Finally, the proposed method's effectiveness was validated using the CWRU bearing dataset and the research group's self-test bearing dataset. Experimental results demonstrate that the proposed method effectively improves cross-domain imbalance sample fault diagnosis accuracy and generalization capability.

Traditional simultaneous localization and mapping (SLAM) systems are typically based on the assumption of a static environment, which often leads to decreased pose estimation accuracy in dynamic scenes. While introducing semantic segmentation can identify dynamic objects, it often incurs significant computational overhead, affecting system real-time performance. To address these issues, this paper proposes an efficient RGB-D SLAM system for dynamic environments. It uses instance segmentation to achieve pixel-level differentiation between rigid and non-rigid objects, avoiding the misidentification of static features by traditional bounding box methods. Combined with a keyframe triggering and optical flow residual change-based mask propagation mechanism, it adaptively updates the dynamic region mask, thereby reducing the computational burden of segmentation. Simultaneously, a model based on adaptively adjusted optical flow residual thresholds using the Fisk distribution is constructed to identify unknown dynamic objects beyond semantic priors. Finally, through a differentiated dynamic feature removal strategy, the system effectively suppresses dynamic feature interference while fully utilizing static features. Extensive experiments on TUM, BONN, and self-collected datasets demonstrate that this method achieves superior localization accuracy and real-time performance compared to existing advanced dynamic SLAM methods in complex dynamic scenes, showcasing its potential value in applications such as robot localization and navigation.

This paper proposes a non-destructive defect inspection technique based on infrared thermal imaging and improved YOLOv5 algorithm to achieve accurate inspection of finned heat sink welding rate, which can be used to improve the quality of finned heat sink defect inspection. Using fin heat dissipation and infrared thermal imaging principles, samples with heat sink quality problems can be efficiently identified, guaranteeing high quality acquisition of samples. By improving the YOLOv5 algorithm applying BiFPN (Bidirectional Feature Pyramid Network) feature fusion method to improve the neck structure of the original network structure, simplify the convolution nodes, add the CBAM (Convolutional Block Attention Module) module to improve the feature extraction capability and inspection efficiency, and optimise the original loss function and prediction frame screening method in order to improve the infrared target inspection accuracy. The experiment verifies that the improved model can effectively detect defective targets under different backgrounds, and the average inspection accuracy (mAP) can reach 90.3%, which makes the model more adaptable and reliable in practical applications compared with traditional target inspection algorithms.

Steel wire ropes are widely used in hoisting, transportation, mining and marine engineering, and their complex braided structure provides excellent tensile strength and fatigue resistance. To overcome the limitations of conventional methods for detecting steel wire rope diameters—particularly in terms of accuracy, robustness, and adaptability to complex working environments—this study proposes a non-contact, real-time detection approach based on super-resolution (SR) technology. The method employs an RGB-Depth camera to simultaneously acquire both visual and depth data of the steel wire rope, and leverages an enhanced super-resolution network (W-EDSR) in combination with a self-supervised deblurring network (SRN-WDeblur) to enhance image clarity and mitigate motion blur. Following this, a semantic segmentation and edge extraction algorithm, augmented with depth information, is applied to derive continuous and smooth rope boundaries. Experimental results indicate that this method achieves 99.35% accuracy for static steel wire rope measurements and maintains 99.24% accuracy when the rope moves at 10 m/s, significantly outperforming traditional approaches and demonstrating strong robustness against motion blur. This research provides a reliable and efficient solution for online monitoring of mine hoisting steel wire ropes, contributing to enhanced intelligence and safety in mining operations.

This paper proposes a method for accurately measuring the angular speed of rotating machinery based on corner transformation and designs a corresponding sensor system. The system primarily consists of a grating and a highspeed camera. By attaching a specific rectangular pattern to the shaft surface, a phase-modulated signal is generated during the shaft's rotation. The instantaneous angular speed (IAS) is then extracted using image feature tracking and corner detection algorithms. Unlike traditional encoders, which rely on resolution, this method theoretically enables angular speed detection with infinite resolution, making it particularly suitable for condition monitoring at low speeds or even near-zero speed. This paper first establishes a rotating shaft model in a simulation platform to simulate the sensor imaging process and validate the method's feasibility through corner detection. An experimental testbed is then constructed to test the sensor under constant and variable speed conditions. Experimental results demonstrate that this method maintains high detection accuracy at extremely low speeds and during dynamic acceleration and deceleration, with the measurement results highly consistent with the motor's setpoints. Further error analysis reveals that the grating attachment accuracy, lighting conditions, and mechanical vibration have some influence on the measurement results, but image preprocessing and feature normalization methods can effectively reduce these errors. Research shows that this method not only has high accuracy and stability, but also has the advantages of simplicity, low cost, and a wide application range. It can provide a new technical approach for condition monitoring, process control, and predictive maintenance of rotating machinery.

The collection of labeled data for transient mechanical faults is limited in practical engineering scenarios. However, the completeness of sample determines quality for feature information, which is extracted by deep learning network. Therefore, to obtain more effective information with limited data, this paper proposes an improved semi-supervised prototype network (ISSPN) that can be used for fault diagnosis. Firstly, a meta-learning strategy is used to divide the sample data. Then, a standard Euclidean distance metric is used to improve the SSPN, which maps the samples to the feature space and generates prototypes. Furthermore, the original prototypes are refined with the help of unlabeled data to produce better prototypes. Finally, the classifier clusters the various faults. The effectiveness of the proposed method is verified through experiments. The experimental results show that the proposed method can do a better job of classifying different faults.

Metallic wire rope is the core component unit in construction, and vision-based nondestructive inspection methods for condition monitoring of metallic wire ropes have progressed in recent years. The lay length of metallic wire ropes contains the stress state and healthy state information of metallic wire ropes, which has important research significance and research value for the force balance guarantee of multirope structures and the safety of buildings. This article summarizes the shortcomings of the existing research of the lay length measurement methods and proposes a measurement method of the lay length of metallic wire ropes via deep learning and phase correlation image analysis algorithm. The segmentation of strand module applies the Mask-RCNN network to segment the strand into Voronoi-like images with high signal-to-noise ratio (SNR). Then, the phase correlation analysis is carried out on the obtained Voronoi diagram to calculate the lay length of the metal wire rope. The experimental results show that compared with the traditional manual method to measure the lay length of metal wire rope, the method proposed in this article greatly shortens the time required for lay length detection, only 0.1045 s. The average detection error of this method is 1.0672 mm, which is far lower than the traditional manual measurement method and the method of directly performing phase correlation image analysis on the metal wire rope. The reliability of the proposed method was validated by varying the lay length of metallic wire ropes using the tension-slack experimental device.

Faced with noise and interference in industrial environments, the need to capture high-frequency and low-frequency signals, the requirements for high resolution and sampling rate of data acquisition systems, and the influence of harsh environments and vibrations, sensors cannot guarantee the stability of signal transmission and the integrity of data. This paper presents the principles and implementation of an innovative sensor system designed to measure the instantaneous angular parameters of rotating shafts at extremely low speeds, approaching zero velocity. The sensor system mainly consists of a sinusoidal modulation pattern and a set of image acquisition devices. The modulation pattern is affixed to the shaft's surface and rotates in unison with it. The camera is used to obtain the characteristic image of the pattern when it rotates with the shaft to calculate the instantaneous angular parameters. When the shaft rotates, the image feature is the phase modulation signal, and the modulation signal can be obtained by image feature tracking and mid-axis feature point extraction methods. The sensor system was experimentally verified using a rotating shaft experiment. The results show that the sensor measurement system is largely consistent with the motor speed setting value. The results show that it has high accuracy and reliability when obtaining low-speed operations with acceleration and deceleration transients.

RV reducer operating conditions are complex, often with low speed, reciprocating, non-integral cycle and other characteristics, coupled with the RV reducer itself complex structure and fault signal instantaneous, so the fault signal sample is small, which brings challenges to the bearing fault diagnosis of RV reducer. Therefore, an improved DNCNN denoising algorithm is proposed to realize the failure mode recognition of RV reducer. The algorithm improves the basic unit module (Conv(3*3) + BN + Relu) in DNCNN network, which can effectively solve the problem that the stacking of convolutional blocks leads to a large number of network parameters, makes the training time long, and most of the training parameters are concentrated in a certain layer of the network. The SE attention module is introduced to avoid the problem that the CNN model cannot effectively highlight the key points when extracting features. At the same time, the Swish activation function is used to replace the Relu activation function, which improves the phenomenon that the latter cannot learn the value of the function input less than 0, and the parameters cannot be updated. Based on improving the DNCNN denoising algorithm, the neural network fault diagnosis method of the main bearing state of the three RV reducers after denoising is studied, and the CBAM attention mechanism module is added to the network to obtain the InceptionV4-CBAM network model. Multiple sets of experimental analysis show that the failure mode recognition rate of the main bearing of the RV reducer under noise interference by the Improved DNCNN combined with InceptionV4-CBAM is 97.3 %.

Accurate gear parameter inspection technology can effectively improve the quality of gear production. In order to obtain more accurate detection results, the quality of the detection image must be controlled. However, there is a lack of optimization of image accuracy in the existing gear size detection methods. Consequently, this article proposes a super-resolution (SR)-based dimensional detection method for small modulus injection molding gears. The method introduces an improved SR algorithm for small modulus injection molding gears, constructs a size detection system, effectively enhances the quality of the image of the measured gear through image processing, and then calculates the value of the measured gear's precision parameters according to the definition of gear precision parameters. The experimental results show that the standard deviation of the bore diameter and the toothed circle diameter are 0.756 and 1.531 mu m respectively, and the repeat errors of the repeated measurements are less than 2 mu m. Therefore, the research results of this article provide a new detection method for related industries, which is of positive significance for improving production efficiency and quality.

This paper proposes a research method for the unfolding of the inner wall view of threaded pipes based on convolutional neural networks. Firstly, an industrial endoscope is used to capture the inner wall view of the threaded pipe. Then, a relationship between the lens position and the imaging of the inner wall view is established. An image correction method based on perspective transformation theory is proposed to correct the distortions present in the inner wall view. Finally, an improved image radial unwrapping algorithm is presented, which combines convolutional neural networks with image registration. The algorithm is extended based on the VoxelMorph framework and performs radial stretching unwrapping of the images to obtain the planar unfolded view of the threaded pipe’s inner wall. Through experimental analysis, the proposed algorithm is compared with traditional SIFT and SURF algorithms. The algorithm shows advantages in terms of RMSE (Root Mean Square Error) and SSIM (Structural Similarity Index Measure). The RMSE value is reduced by 0.09, and the SSIM value is improved by 0.24. This method is suitable for the detection of the inner wall of threaded pipes with diameters ranging from 5 to 10 cm, and it demonstrates good unfolding results.

Angular contact ball bearings have been widely used in machine tool spindles, and the bearing preload plays an important role in the performance of the spindle.In order to solve the problems of the traditional optimal preload prediction method limited by actual conditions and uncertainties, a roller bearing preload test method based on the improved D-S evidence theory multi-sensor fusion method was proposed.First, a novel controllable preload system is proposed and evaluated.Subsequently, multiple sensors are employed to collect data on the bearing parameters during preload application.Finally, a multisensor fusion algorithm is used to make predictions, and a neural network is used to optimize the fitting of the preload data.The limitations of conventional preload testing methods are identified, and the integration of complementary information from multiple sensors is used to achieve accurate predictions, offering valuable insights into the optimal preload force.Experimental results demonstrate that the multi-sensor fusion approach outperforms traditional methods in accurately measuring the optimal preload for rolling bearings.

The composition of duck down is complex and difficult to separate, so at present, in the process of composition detection of duck down, the composition of duck down are recognized by the method of artificial separation, which is inefficient and subjective. In this paper, a duck down recognition method based on the improved YOLOv8 algorithm is proposed, which realizes image enhancement through image preprocessing and reduces the influence of complex background. By improving the YOLOv8 algorithm, the accurate identification of duck down was realized. The lightweight modules GSconv and VoV-GSCSP are used to increase the speed of model detection and improve the ability to detect multiple targets. Add the CBAM module to improve the feature extraction ability and detection accuracy, and improve the model's ability to detect small targets. The experimental shows that the average accuracy of the improved model for the identification of duck down (MAP50 and MAP50-95) is 99.1% and 68.1%, respectively. By comparing the mainstream target detection networks, it is demonstrated that the improved YOLOv8 has the highest detection accuracy and the fastest detection speed, which makes it practical and reliable for detection of down and similar small and multi-target targets in practical applications.

The vibration signals of the bearings of coal mine machanical equipment under the working conditions of strong impact and heavy load show strong transient non-stationary and local nonlinear features. It is difficult to identify the fault features by the classical time-domain statistical analysis method and the global domain transformation method. The traditional order tracking method has the problems of inconvenient equipment installation and difficulty in obtaining instantaneous frequency. The traditional keyless phase order tracking method estimates the instantaneous frequency with low precision under the condition of severe speed fluctuation. This leads to poor fault identification effect. To solve these problems, a new method of bearing fault diagnosis based on harmonic matching compensation and keyless phase order tracking is proposed. Firstly, the time-frequency analysis method based on harmonic matching compensation is used to process the bearing vibration signal and estimate the instantaneous frequency accurately. Secondly, the Vold-Kalman filtering method is used to adaptively extract the harmonic component signal. Thirdly, the Hilbert transform is used to calculate the instantaneous phase of the harmonic. The mapping relationship between the time domain and angle domain is obtained, so as to complete the resampling of the original time domain signal in the angle domain. Finally, the resampled signals are processed by fast Fourier transform (FFT). The fault features of the bearing are identified by analyzing the envelope order spectrum. The simulation and experimental results show that the maximum relative error between the estimated instantaneous frequency and the actual value is less than 1%. The feature order of bearing fault is accurate and obvious, which can effectively diagnose the bearing fault.

The traditional manual detection is difficult to meet the requirements of high speed, many kinds of defects and large change of defect size of optical thin film defect detection. Therefore, an automatic online detection system based on machine vision is proposed, and an efficient and feasible algorithm flow to meet the online detection of optical thin films is given. After preprocessing the original image with the improved mean filter, the Otsu threshold segmentation algorithm is used to achieve fast and accurate segmentation of thin film defects, which improves the efficiency and accuracy of thin film defect feature extraction and recognition. The corresponding experiments were carried out by using the false detection rate and missed detection rate, and the results show that the system is efficient and feasible. At the maximum detection speed of 300 m/min, the missed detection rate and false detection rate are 4.6% and 4.8%, respectively, which meet the production requirements of enterprises.

With the development of the machinery industry and the popularization of shaft parts, the bending and deformation problems of shaft parts in the machining process have become particularly prominent. In order to reduce the bending deformation and improve the yield of parts, shaft straightening machines have become a necessary choice. According to the development status of straightening machines at home and abroad, and based on the consideration of improving the automation and intelligence level of domestic straightening machines, a fully automatic straightening machine measurement and control system has been proposed and developed in this paper based on the LabVIEW platform. It has been proved that the straightening machine can realize parallel operation of data in the working process, with high straightening accuracy and good operation stability.

A torque sensor is one of the most widely used sensing technologies in modern rotating machinery. It is extremely significant in designing the sensing element with high sensitivity and angle measurement accuracy. This study proposes a double-resolver high-sensitivity torsion spring-type torque sensor via the iterative error self-compensation method. For the first time, the torsion spring and the double resolver are used as a sensing element and a measuring part in the sensor, respectively. Here, the iterative error self-compensation method is presented to solve the problems of unequal amplitude and nonorthogonal phase of the excitation signal. Moreover, the problem of zero error is solved. Furthermore, a phase measurement system based on the field-programmable gate array (FPGA) is designed to achieve a high-precision torque measurement. Here, experiments are performed to verify the feasibility and effectiveness of the proposed sensor. The results show that the proposed sensor has outstanding angle measurement accuracy, sensitivity, and other statistical characteristics. Hence, it has a potential application prospect in the field of new-energy vehicles.