Abstract:The transmission of optical signals in atmospheric turbulent channels causes signal fading and light intensity flicker, and the baseband signal cannot be recovered correctly by fixed threshold detection at the receiving end, so adaptive adjustment of the received signal decision threshold is required. Adaptive threshold detection technology can effectively suppress the atmospheric turbulence effect, which is an important means to improve the bit error rate performance of optical wireless communication systems and enhance system reliability. Its detection performance is mainly optimized and improved for the threshold detection algorithm and feedback mechanism. Reviewing the development process of adaptive threshold detection technology, starting from the structure of optical wireless communication system, deriving the optimal decision threshold model for the received signal based on Bayesian maximum likelihood estimation and maximum posterior probability criteria, and the realization of baseband signal demodulation by comparing the received signal with the optimal decision threshold. The typical adaptive threshold detection models based on the minimum mean square error filter, Kalman filter and fading Kalman filter are analyzed, which are suitable for stationary input signals and non-stationary input signals respectively. At the same time, the related research work of Xi′ an University of Technology using high-order cumulants instead of traditional second-order statistics in the field of adaptive threshold detection is introduced. Finally, the future development trend of this field is summarized and foreseen, which can provide some reference for the future research and development of adaptive threshold detection technology for optical wireless communication.

Abstract:A reliability assessment model based on the total probability formula is proposed to address the potential fatigue failure and fracture failure of MEMS acceleration sensors under complex stress conditions. The model accomplishes reliability modeling of the device in vibration, impact, and vibration-impact coupled environments. The model includes the Wiener processes and the homogeneous poisson random processes, which describe the fatigue damage of the device in the vibration environment and the random impact of the device, respectively. Furthermore, the influence of the amplitude of random impacts on the device degradation rate is considered. The correlation between multiple failure modes is reflected by the sudden increase in fatigue damage generated by the device under impact stress. A comparative analysis was conducted to compare the reliability models considering the independent and coupled effects of vibration and impact. The results demonstrate that the reliability model considering the coupled effects of vibration and impact provides more meaningful guidance in the evaluation.

Abstract:Aiming at the problem of deficient demodulation precision in fiber Bragg grating ( FBG) demodulation system caused by wavelength drift in a variable temperature environment, a grating method for FBG sensing system based on Fabry-Perot (F-P) standard for nonlinear wavelength drift calibration was proposed. In this paper, the FBG demodulation system based on the F-P standard is established, the wavelength calibration process based on the nonlinear error compensation of the F-P standard is designed, and the simulation experiment system of the FBG demodulation system in the practical application environment of aerospace is built, and the experimental tests are carried out in the constant temperature and variable temperature environment. The experimental results show that the demodulation stability of the sensor system is within ±0. 6 pm and ±1. 1 pm under constant temperature and variable temperature, and the wavelength drift of the sensor system demodulation decreases from 20. 3 pm to 0. 7 pm under variable temperature. The accuracy and stability of the method are verified, which provides a reference for structural health monitoring in aerospace and other fields.

Abstract:A DV-Hop (distance vector-hop) localization algorithm based on dung beetle algorithm optimization was proposed for the problem of significant localization error of traditional DV-Hop algorithm in wireless sensor networks. Firstly, the dual communication radius was introduced to refine the number of hop nodes, then the average hop size of anchor nodes was calculated using the minimum mean square error criterion, and the mean of the improved average hop size was taken as the average hop size of each unknown node, finally, a weighting factor was introduced to optimize the fitness function, and the dung beetle optimization algorithm was used for coordinate calculation instead of the trilateral measurement method. The simulation results show that the proposed algorithm improves the average positioning error by 55. 69%, 59. 61% and 67. 59%, and the error variance by 52. 41%, 45. 58% and 36. 87% than the classical DV-Hop algorithm, which has good positioning accuracy and better stability.

Abstract:This article proposes a high-precision batch temperature measurement system and testing method for low temperature measurement efficiency of high-precision temperature sensors, the system with the ability testing multiple cores. This article proposes classification method, binary scanning method, and successive approximation method to calibrate the platinum resistance PT100 for the low testing accuracy of platinum resistance PT100. After calibration, the maximum temperature measurement error value in the temperature range of -65 ℃ to 145 ℃ is changed from 0. 412 ℃ to 0. 021 ℃ , and the temperature measurement accuracy is improved by 94. 9%. Simultaneously, the stability, temperature measurement time, and temperature measurement accuracy of the system are modeled and simulated. Firstly, Heat transfer modeling is conducted for the outer wall to the inner wall and convective heat transfer of cold fluid in internal cavity of the thermostatic device respectively. The simulation result shows that the thermostatic device can achieve heat balance in only 75. 2 seconds. Then, the tested circuit heating up, the base heating up, and the calibrated platinum resistance is modeled by comprehensive thermal simulation. The simulation result shows that the temperature measurement accuracy of the system can reach to 0. 016 ℃ . Finally, the external and internal temperatures of several classic high-precision temperature sensors are tested and verified, and the result shows that the system can test temperature sensors with a temperature measurement accuracy of 0. 031 ℃ , which can well meet the temperature measurement requirements.

Abstract:The geological conditions of the slopes on both sides of the reservoir area of the hydropower station are complex and changeable. The stable working condition of the slopes is closely related to the safe operation of the hydropower station. In this paper, the MEMS attitude sensors (gyroscopes and accelerometers) are arranged on the slopes by the array mode at the left slope cover layer of the reservoir area of the Yalong River Guandi Hydropower Station. Using the initial alignment, a real-time slope stability data acquisition system is established to determine the navigation target parameters′ attitude, orientation, velocity and position. The initial values of philosophy, bearing, speed and position are calculated. Then the acceleration and angular momentum under the load system acquired by the MEMS attitude sensors are converted to the navigation system in real-time through the attitude matrix, which is used to calculate the slope change attitude matrix and displacement in real time to carry out experimental research. The experimental results show that applying the MEMS attitude sensor to slope displacement monitoring can achieve real-time acquisition of slope surface displacement data. The displacement data acquisition accuracy can reach the millimetre level to meet the actual needs of engineering. The research results are of great significance to the real-time monitoring and analysis of the stable state of slope surface displacement.

Abstract:In order to reduce the diagnostic complexity and cost of the three-level neutral-point-clamped ( NPC) rectifier, a fault diagnosis method for the open-circuit fault of power switches was proposed. This method was based on the characteristics of switching states. Firstly, the cumulative values of some switching states in half a current cycle were selected as the diagnostic variables, which were combined with the self-adaptive threshold and the phase of currents to realize the faulty bridge arm detection. Then, the outerswitch fault-tolerant control based on reactive current injection was implemented and the threshold was updated. Finally, according to the change of diagnostic variables, the faulty switch was located during fault-tolerant process. The proposed method can realize both singleswitch and double-switch open-circuit fault diagnosis without additional sensors, complicated calculations and diagnostic rules. It has the advantages of simple implementation and low cost. The experimental results verify the effectiveness and robustness of the proposed method.

Abstract:Aiming at the low accuracy of object six-degree-of-freedom ( 6D) pose estimation in scenes with interferences such as illumination changes, distance changes, background clutter, and occlusions, a mixed channel attention module (MCA) is proposed, which combines multi-scale feature fusion and attention mechanisms. Based on MCA, a category-level object 6D pose estimation method (MCA6D) is further constructed. The key steps include object instance segmentation, feature extraction and optimization based on MCA, object model reconstruction based on prior shape, and pose estimation based on point cloud registration. Relevant experiments show that our method achieves 86. 3% (5°2 cm), 73. 4% (5°5 cm) and 39. 2% (5°2 cm), 43. 3% (5°5 cm) mean average precision in the public datasets CAMERA and REAL, respectively, which is ahead of mainstream methods such as NOCS, SPD, and SGPA. At the same time, the practical experiment shows that the proposed method can accurately estimate the 6D pose of the object in scenes with interference, such as illumination changes, distance changes, background clutter, and occlusions.

Abstract:In multi-user multi-channel cognitive radio networks for the internet of things (CR-IoT), a spectrum co-sensing-allocation strategy based on edge computing is proposed. To evaluate the performance of the proposed strategy, a performance evaluation system (PES) capable of quantifying the quality of service (QoS) of cognitive users and the interference to primary users is developed. In the PES, a Markov model is established to describe the CR-IoT system state based on queuing theory. The performance metrics of heterogeneous cognitive users with configurable parameters can be analyzed independently. Thus, various spectrum sensing-allocation strategy can be evaluated using the proposed PES. Numerical results show that the proposed edge computing strategy outperforms central computing strategy in QoS of cognitive users and the quantified interference to primary user. It proves that the PES can help regulators design various spectrum sensing-allocation strategy.

Abstract:With the wide application of lithium-ion battery systems in electric vehicles, the safety issue caused by short-circuit fault of battery pack is becoming more serious. Therefore, the studies on state monitoring of battery pack and fault diagnosis are receiving more attention. To deal with the issues of low generality, poor anti-interference capacity and critical inconsistency of battery pack existed in non-model-based fault diagnosis methods, a short-circuit fault diagnosis method based on statistical analysis and density clustering is proposed for battery packs in this paper. Firstly, the fault information of battery pack is extracted by using the relative entropy of kernel density estimation (KDE) and correlation coefficient, based on a forgetting mechanism. The fault information is used to identify the changes of batteries’ voltage and temperature caused by short-circuit fault. Then, the short-circuit battery can be automatically identified by adopting the density-based spatial clustering of applications with noise (DBSCAN) algorithm. The robustness of the proposed method is validated under conditions of noise interference and serious inconsistency. Furthermore, the effectiveness of the proposed method is verified under different short-circuit degree with 1, 5 and 10 Ω short-circuit resistors, and the accuracy of short-circuit fault diagnosis can reach 92. 17% in the case of a 10 Ω short-circuit resistor. By comparative analysis, the results show that the proposed diagnosis method can effectively detect and locate short-circuit batteries, and the more severe the fault, the shorter the diagnosis time required.

Abstract:At this stage, the image deep learning algorithm cannot detect the chronological process problem. In this paper, the artificial assembly process of the mountain board assembly of knitting machinery is studied, and the MS-RetinaNet object detection algorithm is proposed. Using the idea of natural language processing for reference, the Swing-Transformer structure is introduced to retain the hierarchy of CNN structure, make up for the lack of high-level semantic information fusion in CNN structure, and enhance the ability to learn overall and details. The improved GIoU Loss is used to increase the judgment factor formula, mitigate the impact of loss calculation degradation, and optimize the regression effect of the bounding box. According to the multi-scale target parameters, the best anchor frame ratio is adopted to improve the recall rate and detection accuracy. The chronological detector is designed to enable the algorithm to distinguish the sequence and logical relationship of the target. The experimental results show that the algorithm AP can reach 90. 3%, which is more than 2% higher than the current mainstream algorithm. The detection speed of a single image is about 46 ms, meeting the chronological detection requirements of the process flow, and the overall performance is superior.

Abstract:Aiming at the problems of solid thread gauges including difficult to process, high cost, easy wear and cumbersome verification procedures, a digital measurement model establishment and parameter calculation method of thread gauge based on three-dimensional point cloud was proposed. In order to realize the digitization of the physical thread gauge. First, the 3D digital model was obtained, and the key parameters of the model such as large diameter, middle diameter, small diameter, pitch and tooth profile angle were obtained. Secondly, the single error and comprehensive error of the model are analyzed, and the relative error of each parameter is less than 5%. Finally, the uncertainty and extended uncertainty of each parameter of the digital model are calculated, which proves the accuracy of the digital measurement model of thread gauge. The results provide a research basis for the digital development of thread gauge and have reference value for the digital transformation of measurement industry.

Abstract:A semantic network-based network search algorithm is proposed to address the problems of slow planning speed and high memory occupation of raster map-based path planning techniques in the face of large maps and high-resolution maps. Firstly, a semantic partitioning network is used to pre-sample the raster map, secondly, the optimal path range is formed by widening the optimal path through imagery expansion to improve the robustness of the algorithm, finally, the feature map of the semantic network is used to guide the planning of the search algorithm, which speeds up the path planning of the high-resolution raster map. Experimental simulations show that the network search algorithm reduces the time by an average of 72. 5%, the number of traversal points by an average of 51. 6%, and the path length by an average of 0. 73% compared to the traditional search algorithm, and the network search algorithm can effectively speed up the path search and reduce the memory occupation.

Abstract:Aiming at the problems of the northern goshawk optimization algorithm (NGO), such as low convergence accuracy and easy to fall into local optimum, an improved northern goshawk optimization algorithm (INGO) is proposed and applied to the fault diagnosis of photovoltaic array. Firstly, circle mapping, adaptive weight factor and Levy flight strategy are used to improve the INGO. Combined with Gaussian detection mechanism and hybrid kernel extreme learning machine ( HKELM), the INGO-HKELM fault diagnosis model is built. Secondly, the INGO algorithm is compared with the NGO, the particle swarm optimization algorithm ( PSO), and the whale optimization algorithm (WOA) on the test functions, which shows that it has advantages in optimization ability and stability. Then, the operating characteristics of photovoltaic arrays under different operating states are analyzed, and a 5-D fault feature vector is proposed as the input of data. Finally, the four algorithms are used to optimize the kernel parameters of HKELM and achieve fault classification. The results show that the proposed method can accurately detect abnormal states of photovoltaic modules, and the accuracy of INGO-HKELM model reaches 93. 74%, which verifies the effectiveness and feasibility of the proposed algorithm.

Abstract:In order to solve the problems of the ORB image feature detection algorithm under non-uniform illumination, such as overly clustered feature points and low accuracy of feature matching, we propose an efficient and high-precision illumination adaptive ORB image feature matching algorithm. The oFAST feature points of the image to be measured are extracted using the adaptive threshold, and the number of feature points in the low illumination or high exposure area is further increased through the uniform distribution of the optimized quadtree decomposition method. Then, feature matching is performed according to Hamming distance, and the improved RANSAC algorithm is used to eliminate mis-matching, so as to improve the matching accuracy of the feature points in the ORB algorithm. The experimental results show that for data sets with obvious illumination changes, compared with ORB, MA, Y-ORB and SORB algorithms, the average feature distribution uniformity of our proposed algorithm is improved by 13. 1%, the feature extraction time is saved by 26. 3%, and the comprehensive evaluation index is improved by 18. 5%. It can efficiently complete feature matching under complex environment changes, and has strong application value in the fields of target recognition and 3D reconstruction.

Abstract:In order to solve the problem that local specular reflection is easy to occur during the measurement of objects with strong reflection surface, which affects the measurement accuracy. The digital micro mirror is added to the structural light measurement optical path and the measurement system is designed and built, including the digital micro mirror, CCD camera, projector and other devices. The matching of the system, the parameter calibration of camera and projector and the phase calculation are completed. The coordinate mapping relationship between the digital micromirror device unit and the camera pixel unit is established by using the sparrow search algorithm to optimize the BP neural network. The mapping error is 0. 583 pixels. An adaptive mask generation method based on PID controller is proposed, and the measurement experiment of the measure block with strong reflective surface is carried out. The experiment shows that the method can effectively reduce the gray level of the overexposed area, and achieve high dynamic range imaging. The proposed method can provide theoretical support for the three-dimensional measurement of strongly reflective surfaces.

Abstract:In order to improve the control performance of single-phase five-level rectifier, a dual closed-loop integrated control method based on non-delay power observer was studied. In this method, a virtual current signal reconstruction algorithm is proposed and combined with the improved generalized integration algorithm to construct a non-delay power observer, which is applied to the model predictive power control of the inner loop of the rectifier to observe the required power in real time, while the linear distraction control is used instead of the traditional proportional integration algorithm by linear distraction control in the outer voltage loop. The experimental results show that the time required for the control method to retrace the given value after the power mutation is shortened by 5~ 7 ms, and the voltage fluctuation on the DC side is reduced by 4. 4% and 4. 8% under the grid voltage drop and load disturbance conditions, respectively. Compared with the traditional method, this method not only reduces the overshoot of rectifier, improves the dynamic performance of the predicted power control of the inner loop model, but also effectively enhances the anti-interference ability of the voltage outer loop.

Abstract:The proportion of armor clamp rust in aerial images of power transmission lines is rich in details and irregularly distributed. To overcome problems such as local information loss, low accuracy, and slow speed in the segmentation detection process, a DeepLabV3+- based semantic segmentation model for armor clamp rust is proposed. The backbone network is replaced with a lightweight improved MobileNetV3 network to speed up computation, and an adaptive feature pyramid (AFP) structure is proposed to merge multiple scales. A feature fusion atrous spatial pyramid pooling ( FEF-ASPP) structure is proposed, combined with the FRN layer to strengthen pixel relationships without reducing resolution. Finally, the loss function is optimized to improve the effectiveness of the operator. Experiments show that the mIoU and mPA reach 87. 15% and 96. 64%, respectively, which is an improvement of 3. 09% and 4. 29% compared to the original model. The parameter quantity is only 48% of the original model, and the inference time is only 15. 94 ms, reducing the requirement for device computing power and achieving high-efficiency, high-precision, and lightweight segmentation detection of armor clamp rust in power transmission equipment.

Abstract:For the lack of detection accuracy for human keypoints, it is improved on the basis network of KAPAO (keypoints and pose as objects). The generalization of network is improved by the enhance data method of PoseTrans ( pose transformation); for the lack of characteristic fusion capabilities, the BiFPN (Bi-directional feature network) module is designed to fully integrate different semantic characteristic to improve the integration ability of deep semantics information and shallow semantic information; the adaptive expansion convolution module is designed to adaptive fusion different expansion rates of output branch during the network output phase, it effectively obtains the global information of the image; in order to retain the optimal key point prediction box, the traditional NMS is replaced by SDR-NMS ( soft DIOU relocation non-maximum suppression ) during the post-processing part of the network. The experimental results show that the AP score was increased by 4. 8%, the AP was 68. 6%, and the detection speed was 19. 1 ms. The accuracy and detection speed of network have better performance.

Abstract:Accurate prediction of wind speed is of great significance for safe operation and efficient power generation of wind farms. Aiming at the inherent defects of the single decomposition strategy used in existing literatures in wind speed prediction and the unstable effect of the optimized prediction model, a hybrid prediction model combining two-stage decomposition and iJaya-ELM is proposed. First, ICEEMDAN decomposition is performed on the original wind speed sequence, and 12 components are obtained, and reconstructed into high frequency terms, middle frequency terms and low frequency terms based on the permutation entropy. Then, the high frequency term is filtered by singular spectrum decomposition to remove the sequence noise. An improved Jaya algorithm, iJaya, is proposed to obtain the optimal connection weights and thresholds of ELM. Finally, the predictive results of each component are linearly integrated to obtain the final results. The model is validated by wind speed data of wind farm in Gansu province of China, and its robustness and universality are tested by wind speed data of Xinjiang region. The experimental results show that the iJaya algorithm is of strong optimization accuracy and stability, and the two-stage decomposition can deeply excavate the characteristics of wind speed series. The hybrid model can effectively improve the wind speed prediction accuracy, and the average absolute error and mean square error are 0. 067 9 and 0. 134 5, respectively.

Abstract:In the double-sided shearing process, the steel plate alignment process requires manual visual observation of the laser beam allowance, which is complex in operation and subjective judgments that affect data accuracy. Therefore, in this paper, an automatic alignment system for double-sided steel plate shearing based on machine vision is designed, which relies on multiple sets of area array cameras distributed along the roller table to collect the status data of the steel plate on the roller table. Using on-site measurement data and the target width of the steel plate, two virtual cutting lines are calibrated, eliminating the dependence on traditional auxiliary laser lines. At the same time, a cascaded steel plate object extraction model is adopted in the system, and the step-by-step extraction idea of “rough first and then fine” is adopted to improve the accuracy of steel plate edge detection. The movement distance is converted based on the relationship between the steel plate contour position and the virtual shear line position, thereby controlling the magnetic centering device to complete the steel plate centering process and improving the automation of the double sided shear process. The actual application results show that the system has a measurement error of less than 5 mm for the width of steel plates, and an automatic control centering error of less than 10 mm, meeting the automatic control requirements of enterprises.

Abstract:Aiming at the problem of poor generalization ability of fault diagnosis of traditional deep learning network model under variable working conditions, a fault identification method based on the fusion of transfer learning bidirectional long short memory network and attention mechanism ( TLBA) is proposed. Divide the original fault data into source domain and target domain, and construct a bidirectional long short-term memory network (BA) model that integrates attention mechanisms, and then use this model to learn source domain data features. Finally, transfer learning is used to further optimize and adjust the network parameters of the BA model by learning the data in the target domain, and finally the fault classification identification model in the target domain is obtained. Taking the aircraft wing beam fault as an example, the results show that compared with the traditional fault diagnosis method BiLSTM-Attention, the comprehensive evaluation index F1-score of this method is improved by 3. 4%, and the average fault diagnosis accuracy is above 91%. At the same time, the fault classification results under variable operating conditions are relatively stable.

Abstract:As an indispensable key component of power transmission lines, power fittings provide a guarantee for stable power transmission. Once the electric power fittings have defects, it will bring huge hidden dangers, causing damage to transmission facilities or even large-scale power failure, affecting people’ s production and life. The traditional power transmission line maintenance mainly depends on manual on-site maintenance, which is not only dangerous, but also difficult to detect. The continuous progress of AI recognition technology provides a better method for the defect recognition of electric power fittings. At present, the target recognition accuracy of Faster R-CNN is high, but it is relatively low for small target objects such as screws. Firstly, the features are extracted and marked by the double feature fusion operator, then input into the improved Faster R-CNN model with the introduction of mixed attention mechanism for feature re extraction. The features with high coincidence degree are fused, and the defects are classified and recognized, which can effectively identify the screws in the small power fittings. The experiment shows that the improved Faster R-CNN based on dual feature fusion in this paper has obvious improvement effect compared with the traditional Faster R-CNN and YOLO. The average accuracy of the model is improved by 5%, and the average accuracy is improved by 11%, which also ensures the real-time performance of the algorithm identification. It has a good detection effect on small electrical fittings such as screws.

Abstract:The sag of conductor and ground wire is one of the key indicators for the construction quality and safety operation of the overhead transmission line. Aiming at the shortcomings of current sag measurement method’s accuracy and convenience, a mathematical model is proposed based on laser ranging and grating angle measurement technology for the sag measurement of the line. On the basis of this model, a data quality evaluation optimization algorithm which the similarity measure combined with the Newton interpolation method is put forward to realize the data set compensation and sag calculation for the case of insufficient robustness of the single conductor’s sag measurement of the multi-split line. Compared with point cloud extraction power line method, only a small amount of measurement data is needed for the proposed method to fit the line model and calculate the sag value. In order to verify the proposed method and compare with current sag measurement methods, a 220 kV double-split line is used in a test, and the result shows that the maximum sag error rate of the optimized data is only 1. 47%, which proves that the proposed method’ s measurement accuracy can meet the requirement of engineering sites, and can improve the security and efficiency of the overhead transmission line’s sag measurement.

Abstract:The corrosion detection of main ground electrode in underground coal mine is crucial to the safety of the staff. Traditional manual visual inspection detection by transporting the electrode overground is time-consuming and the damage magnitude cannot be confirmed. Therefore, a new online corrosion detection method based on ultrasonic SH wave echo characteristics is proposed. Firstly, the dispersion equation of SH wave in the liquid-immersed plate structures is derived. The zero-order SH wave with small dispersion and long propagation distance is selected as the detection excitation signal, and the finite element model of liquid-immersed plate structure is constructed to determine the optimal ultrasonic excitation frequency. Secondly, the coupled flow-solid-acoustic multi-physical field detection model of the main grounding electrode in coal mines is constructed and the corrosion detection ability of ultrasonic SH0 wave on the main ground electrode is studied by simulation, then the echo signal characteristics under 1 ~ 5 mm corrosion depth and 5 ~ 25 mm corrosion radius are analyzed. Finally, the non-destructive testing system of the main ground electrode in coal mine is constructed, and experimental verification is carried out. The experimental results prove the feasibility of the proposed method, the corrosion location error is 5. 64% on a 500 mm×375 mm×5 mm main grounding plate, the corrosion signal amplitude is positively correlated with the magnitude of corrosion defects. The research provides an effective method for corrosion location and damage assessment of underground coal mine main ground electrode.

Abstract:Aiming at the problem that the screen data of roll grinder can only be obtained by manual transcription, we designed a method of automatic identification and recording of key parameters of roll grinder screen. The specially designed smart camera is installed above the CNC screen of the grinder, and the camera structure is designed with the angle of 45° “L”, which can take photos of the CNC screen without affecting the work of the master. Firstly, the screen image is registered and corrected by edge positioning and perspective transformation. Secondly, grinder parameters in the image were identified by the trained YOLOv5 model. Finally, the key parameters of the grinder are imported into the database to complete the real-time recording and transmission of the parameters, so as to provide timely and accurate key equipment parameters for the adjustment of related subsequent production processes. In addition, in view of the common Moire pattern phenomenon in screen images, the design of polarization window and Moire pattern removal algorithm is combined to effectively filter Moire pattern, which significantly reduces the influence of Moire pattern on the recognition accuracy. Since the system has been running for half a year, the recognition rate of grinding machine screen data has exceeded 99%, which significantly reduces labor intensity and manual error, and improves productivity.