51 results about "Fault detection algorithm" patented technology

The invention relates to the technical field of operation and maintenance of data centers, and discloses an intelligent operation and maintenance monitoring method and system of a data center, and the method comprises the steps: building a hierarchical monitoring frame, carrying out the hierarchical data collection and standardization processing of three layers, namely, physical equipment, virtual resources and application services, and constructing an interlayer data sharing model; performing anomaly scanning on the running state of each layer by using a preset fault detection algorithm, triggering an alarm once an anomaly is found, and determining a propagation path influenced by a fault through data backtracking; and then analyzing a causal relationship between anomaly and cross-layer data by using an association rule mining algorithm, performing multi-dimensional verification in combination with a decision support rule base, and finally generating a fault positioning conclusion and a repair priority. In addition, response scripts can be automatically called for parameter adjustment, rapid positioning and automatic repair of faults are achieved, and the operation and maintenance efficiency and reliability of the data center are effectively improved.

The invention relates to the technical field of pumped storage, and discloses a state detection method for a water energy storage unit, which comprises the following steps: S1, deploying vibration, temperature, electrical and pressure sensors on the water energy storage unit, collecting multi-dimensional operation data, and carrying out noise reduction, standardization and time sequence alignment; s2, multi-dimensional features of the data are extracted through tensor decomposition, local space information is acquired by using a convolutional neural network, and a time sequence relation is modeled in combination with a Transform model; and S3, constructing a sensor topological structure through the graph convolutional network. The multi-source data fusion technology is adopted, data of vibration, temperature, electrical and pressure sensors are processed in a unified mode, time sequence prediction and an intelligent fault detection algorithm are combined, comprehensive health monitoring of the water energy storage unit is achieved, the detection precision is improved, meanwhile, potential faults can be found in time, and compared with data monitoring of a single sensor in a traditional method, the method has the advantage that the monitoring accuracy is improved. And the problem of inaccurate identification caused by large and complex data volume is obviously solved.

The invention provides a dynamic fault-tolerant control method based on an expected function safety mechanism, and the method comprises the steps: 1, selecting the model of a sensor, and enabling the sensor to be distributed at the periphery of a vehicle body; 2, building a decision algorithm model, carrying out environment scene recognition, planning a driving path and speed according to data obtained by a sensing layer, evaluating safety risks of different decisions, and carrying out judgment and decision making on various conditions; 3, control interfaces of a vehicle power system, a steering system and a braking system are established, and the electronic control unit converts a decision control instruction output by the decision algorithm model into actual vehicle operation through the control interfaces; 4, performing an expected function safety mechanism, performing real-time monitoring on state parameters of the vehicle, and analyzing detection data through a fault detection algorithm; 5, implementing a fault tolerance strategy; and 6, continuously monitoring the actual driving state and the system response condition of the vehicle, and adjusting the fault tolerance strategy.

The invention discloses a suspension system sensor fault detection algorithm, which comprises the following steps: establishing a suspension system model according to the motion of a real part of a suspension system; designing a state observer based on the suspension system model; acquiring and processing a sensor signal to obtain a residual value; and comparing the residual value with a threshold value to obtain fault information of the suspension system sensor. According to the fault detection method based on the algorithm, the change of the system state can be sensed earlier, and an early warning signal is sent out at the initial stage of sensor fault occurrence. The magnetic levitation system is widely applied to scenes of rail transit, air transportation, medical care and the like, the algorithm provides guarantee for normal operation of the system, meanwhile, maintenance personnel provide sufficient time for checking and repairing a fault sensor, further deterioration of faults is effectively prevented, safe and stable operation of the scenes where the magnetic levitation system is applied is ensured, and the service life of the magnetic levitation system is prolonged. And the safety and the reliability are improved.

The invention discloses a fault detection method for a low-voltage AC system containing a distributed power supply based on SOGI and wavelet, and the method comprises the steps: carrying out the filtering of a fault signal through a second-order generalized integrator SOGI, generating an orthogonal signal, and calculating the amplitude; performing wavelet packet decomposition on the amplitude signal, and extracting a high-frequency detail component as a fault feature; and dynamically adjusting a detection threshold according to the permeability of the distributed power supply, and judging a short-circuit state based on the high-frequency detail component. According to the method, alternating current side short circuit fault detection under different fault types under different direct current system source load access conditions is realized, and the rapidity of a fault detection algorithm in a high-permeability scene is improved. The method is beneficial to improving the efficiency, reliability and universality of short-circuit fault identification.

The invention provides a solution for integrating a fault arc detection function into an existing intelligent electric energy meter, and aims at solving the problems that an independent fault arc detection device is low in popularization degree, complex in installation, high in cost, poor in compatibility and the like. The intelligent electric meter comprises a data acquisition module, an electric energy metering module, a fault arc detection sub-module, an A / D conversion module, a storage module, a communication and remote control module, a clock module, a display module, a key module, a power supply module and a main processor module. The data acquisition module acquires voltage and current signals through a voltage transformer and a current transformer, converts the voltage and current signals into digital signals, and extracts high-frequency characteristics and waveform distortion parameters. And the communication and remote control module realizes closed-loop control of fault alarm, user interaction and algorithm optimization. All the modules are coordinated to work through the main processor, and according to a real-time detection result, the following operations are automatically triggered: pushing alarm information; controlling a circuit breaker to cut off a power supply according to a preset threshold; and dynamic upgrading of the fault detection algorithm is completed based on an updating instruction issued by the remote management platform.

The invention provides a whole wind turbine generator bolt monitoring system. The complete machine bolt monitoring system for the wind turbine generator comprises a sensor failure monitoring module, a plurality of high-precision flange bolt gap sensors installed at a fan blade flange and a tower flange, a multi-channel flange bolt gap acquisition instrument used for monitoring flange gaps in real time, and a monitoring module used for receiving monitoring signals output by the flange bolt gap sensors, and data acquisition is carried out. According to the whole wind turbine generator bolt monitoring system, the sensor failure monitoring module and the fault prediction module are introduced, so that the problem of alarm omission caused by sensor failure or failure in the prior art is effectively solved. According to the system, the state of the sensor can be monitored in real time by utilizing a fault detection algorithm based on a model, and an early warning signal can be sent out in time when abnormity occurs, so that blind spots of a traditional system are avoided, and the operation safety and reliability of the wind turbine generator are improved.

The invention discloses a nonlinear system robust residual error multi-fault detection and reconstruction method based on a sliding-mode observer, and solves the problems that an actuator fault and unknown input disturbance in a servo driving system are closely related and mutually coupled, and a fault detection algorithm is insensitive to identification of a first-degree fault, so that fault report failure or misjudgment is caused. According to the method, a general nonlinear system is transformed into two subsystems with independent fault influence conditions based on global differential homeomorphic transformation for the first time. Equivalently transforming reverse channel disturbance to forward channel disturbance by using a filter, designing a sliding mode state observer according to different actuator fault types, and respectively extracting fault information and unknown input disturbance information of a servo driving system in the two subsystems, thereby realizing decoupling of the fault and the unknown input disturbance. And the sensitivity of fault signal detection and the accuracy and robustness of fault diagnosis are effectively improved.

The invention discloses a gray fault processing method and device, a medium and a program product, and relates to the technical field of detection, and the method comprises the steps: collecting the performance data of all server nodes; according to the performance data of all the server nodes and the graph neural network model, detecting the server nodes with gray faults in the server cluster; in response to the grey fault of any server node in the server cluster, verifying the grey fault of the first server node through a single-node grey fault detection algorithm; determining a target migration server node through a server node health scoring mechanism and a topology model of a network link in response to the fact that the gray fault verification of the first server node is correct; and migrating the calculation task of the first server node to the target migration server node. According to the method, the gray fault in the server cluster is detected, the gray fault detection speed and accuracy are remarkably improved, the requirement of manual detection is reduced, and the reliability of the whole system is improved.

The invention provides a DC series arc fault detection method for a photovoltaic system. The method comprises the following steps: analyzing the influence of impedance on fault detection; analyzing the influence of switching noise; designing a resonant filter; and designing an arc fault detection algorithm. The resonant filter designed by the invention can effectively amplify an arc fault signal, so that the FFT amplitude change of the fault signal is more obvious, the normal signal and the fault signal are easier to distinguish, and the detection precision of the direct-current series arc fault is improved; and meanwhile, the arc fault and the normal condition can still be distinguished under the condition of large impedance, and the influence of the impedance of the inverter and the converter on the fault detection capability is reduced. The designed fault detection algorithm can effectively work in a system comprising a direct current optimizer and different commercial inverters, can stably realize a fault detection function, is not obviously influenced by system configuration and working condition change, and realizes the universality of the detection method.

The application discloses a kind of underground pipeline detection method and system based on multiple information sources, which comprises the following steps: pipeline detection system obtains the historical information of regional pipeline, arranges pipeline detection sensor;Establish detection sub-region, obtain pipeline characteristic parameter;Design regional fault detection algorithm, obtain health center point, calculate regional fault coefficient, determine fault sub-region;Pipe robot is put into fault sub-region, design adaptive fault point detection algorithm, detect fault point, record fault point position;Output pipeline detection result, notify pipeline inspection personnel;The scheme realizes the fault detection of underground pipeline by using multiple information sources, improves the detection efficiency by using the historical information of pipeline and establishing the arrangement scheme of pipeline detection sensor, divides the entire pipeline by establishing sub-region, detects each sub-region separately, improves the detection efficiency and accuracy of pipeline fault, and reduces the safety risk caused by pipeline fault.

This invention proposes a multi-stage braking elevator safety clamp control method and system, belonging to the field of elevator braking technology. It includes a base mounted on the elevator car; an elevator guide rail mounted inside the elevator shaft and extending vertically; two sets of synchronous servo electric telescopic mechanisms, positioned opposite each other and spaced apart on the base, located on opposite sides of the elevator guide rail's extension direction; each set of synchronous servo electric telescopic mechanisms has at least two safety clamps at its output end; an acceleration sensor for acquiring the elevator car's acceleration signal; and an intelligent control unit communicatively connected to the acceleration sensor and the two sets of synchronous servo electric telescopic mechanisms. Based on a dual-branch fault detection algorithm using built-in dynamic convolutional branches and improved LSTM branches, when a fault is detected in the elevator, the control unit activates the two sets of synchronous servo electric telescopic mechanisms to perform staged braking of the elevator car when the elevator's acceleration exceeds a set threshold.

The invention discloses a servo motor driver parameter management method and system, and relates to the technical field of motor parameter management, and the method comprises the steps: initializing a parameter database, and outputting a database initialization result; executing a state machine control flow based on a database initialization result; according to the state information of the state machine, executing a corresponding parameter synchronization strategy and outputting a parameter synchronization state; based on the parameter synchronization state and the state machine state information, performing hierarchical error detection and processing, identifying abnormity through a fault detection algorithm based on a time window, performing fault processing according to a hierarchical recovery strategy, and outputting a system health state; and executing system diagnosis and performance optimization according to the system health state and the parameter synchronization state, performing read-write optimization through the parameter cache layer, and implementing predictive maintenance to complete parameter management. Through unified storage and management of data parameters, the reliability of parameter management of the servo motor is improved, and the system has the advantages of high fault identification precision and remarkably improved system stability.

The invention relates to an AGC hydraulic system fault detection method and computer equipment thereof. Data processing modes of a feature extraction method are different (model selection of a preprocessing step and a data processing step). Firstly, candidate measurements needing to be further researched are selected by using typical fault characteristic parameter analysis; second, the selected measurements are analyzed using a feature extraction method. The step aims to extract hidden features so as to detect internal faults to the greatest extent; and finally, the classification efficiency of the fault detection algorithm is evaluated by studying the detection rate of the proposed architecture. According to the method, the detection rate of the accuracy of the AGC hydraulic system is improved, and quick diagnosis and positioning of common faults of the AGC hydraulic system of the automatic rolling mill are realized to a certain extent.

The invention belongs to the technical field of wind turbine generator fault detection, and particularly relates to a PCA-SPE-CUSUM algorithm-based wind turbine generator early fault online detection method, program and equipment and a storage medium. According to the method, principal component analysis and a square prediction error cumulative sum algorithm are applied to monitoring data of the SCADA system of the wind turbine generator set, detection of early faults of the generator system of the wind turbine generator set can be achieved, and early warning can be given earlier than that of the SCADA system. According to the method, the problems that an SCADA data acquisition system is multiple in variable and large in data dimension are effectively solved through principal component analysis, the operation cost is reduced, and the fault detection efficiency is improved. The fault amplitude estimation model deduced based on the fault detection algorithm provided by the invention can realize amplitude estimation of the generator fault of the wind turbine generator, and is helpful for evaluating the severity of the fault.

The invention discloses a wind power plant primary frequency modulation method and an energy management platform, and relates to the technical field of wind power energy. The wind power plant primary frequency modulation method comprises a step following mode and a random disturbance mode. And when the step following mode is lowered, the active power set value of each wind turbine generator is adjusted based on a three-dimensional self-optimization algorithm, and accurate power regulation and control are realized. And during up-regulation, a falling prevention and potential fault sensing algorithm is adopted to pre-determine large-amplitude power falling or potential faults caused by energy over-support caused by wind speed abrupt change of the wind turbine generator, and a strategy of reducing an active power set value of the corresponding wind turbine generator in advance or directly quitting primary frequency modulation is adopted to avoid large-amplitude power falling and off-grid risks. In a random disturbance mode, aiming at the problem that response is not timely due to frequent fluctuation of a control target value, active power of each wind turbine generator is flexibly adjusted based on an adaptive anti-disturbance algorithm, and large control errors and integral electric quantity assessment risks due to the fact that the response is not timely are reduced.

The invention provides an industrial process fault detection cloud system based on LoRa and PCA, and belongs to the technical field of industrial process fault detection. The technical problems that an existing wired system is complex in wiring and limited in communication distance, and a conventional wireless system is short in transmission distance, single in function, lack of fault detection and power supply guarantee and insufficient in simulation verification capacity are solved. The cloud system comprises a LoRa node plate embedded with a PCA (Principal Component Analysis) fault detection algorithm, a LoRa gateway plate, an industrial process simulation platform and a cloud platform, the industrial process simulation platform, the LoRa node plate and the LoRa gateway plate are sequentially connected through LoRa wireless communication, and the LoRa gateway plate is connected with the cloud platform through 4G communication. According to the invention, the dual functions of actual data acquisition and semi-physical simulation are realized, the PCA algorithm is embedded in the node plate, the industrial process is monitored, the fault can be quickly identified, and the fault detection effect is achieved.

The present application belongs to the technical field of fault identification, and particularly relates to a low-voltage series arc fault identification method based on multi-frequency electrical quantity amplitude-phase combination. The steps include: each downstream monitoring point real-time monitors the voltage waveform distortion rate of adjacent cycles, and judges whether it is greater than or equal to the set distortion rate threshold; if yes, the fault detection algorithm is started; the voltage data of the downstream monitoring point and the corresponding upstream monitoring point are aligned asynchronously and in phase, differentiated to obtain the differential voltage, low-pass filtered, and the voltage characteristic energy is calculated; the minimum point of the voltage characteristic energy is obtained, the voltage energy peak values in the left and right windows are obtained and the corresponding phases are calculated; the peak threshold and the fault sensitive domain are determined, and the counting accumulation is performed; if the counting accumulation exceeds the counting threshold, it is determined as a fault. The present application can improve the sensitivity, accuracy and anti-interference ability of the power grid side fault detection.

The invention relates to a distribution network short-circuit fault rapid detection algorithm based on voltage and current synchronous trigger detection. The method comprises the following steps: S1, designing a group of multi-scale low-pass filters, and limiting the highest frequency of voltage and current signals by using the low-pass filters; s2, filtering the filtered voltage and current signals by using a multi-scale low-pass filter, and calculating a filter output difference value between adjacent scales as an abrupt change detection quantity; s3, finding an extreme value of the mutation detection quantity, and calculating an average value of absolute values in specified window widths on the left and right sides of the extreme value; s4, if the short-circuit fault detection is started, starting from the sudden change position of the voltage and the current, using a Savitzky-Golay differentiator to calculate 0-order, 1-order and 2-order normalized differentiations of the voltage and the current in the window after the sudden change position; s5, if the short-circuit fault is detected at the short-circuit fault detection moment but not at the starting moment, judging whether the time difference between the sudden change points exceeds a threshold value or not; if so, detection of the short circuit fault is stopped.

The application relates to the field of fault detection, and provides a bearing fault detection algorithm and system. An original vibration signal is decomposed through a local mean decomposition algorithm, and an effective PF component is screened out by using a Pearson correlation coefficient, and then signal reconstruction is carried out, irrelevant noise signal interference is removed, feature extraction of a bearing degradation process is realized based on a product form of a harmonic noise ratio and a root mean square value, the sensitivity of the harmonic noise ratio to periodic impact is fully utilized, and the problem that the root mean square value has low sensitivity to initial bearing degradation is avoided; a state space model of different bearing degradation processes is constructed based on a Paris model and a Foreman model, compared with a single model prediction model, the prediction precision of RUL is more effectively improved; regularization of a particle filtering resampling process is realized based on an Euclidean distance, the diversity of particles is effectively improved, the particle depletion phenomenon of basic particle filtering is avoided, and the filtering estimation precision of a nonlinear state is improved.

The invention belongs to the technical field of anomaly detection of state information of a control rod driving mechanism, and particularly relates to an anomaly detection method of the state information of the control rod driving mechanism. Acquiring coil current historical data of the control rod driving mechanism; the data is cleaned and normalized; constructing an SAE model, inputting the coil current normal data of the control rod driving mechanism into the SAE model for feature learning, learning the original features of the coil current data of the control rod driving mechanism, and enabling the SAE model to learn a normal operation model of the control rod driving mechanism; coil current data of the control rod driving mechanism in actual operation are put into the model to obtain a reconstruction error; and constructing a random forest algorithm, taking the reconstruction error as the input of the algorithm, and completing the fault detection of the coil current of the control rod driving mechanism. According to the invention, long-sequence and periodic abnormal events can be processed; compared with a traditional fault detection algorithm, the provided SAE and random forest combined method has higher optimization capability and better detection effect.

One example includes a wind turbine yaw control fault detection system. The system includes current monitors that are each configured to monitor a current amplitude of a respective one of a plurality of yaw motors of a wind turbine and to generate a current signal that is indicative of the respective current amplitude. The system further includes a processor to compare the current amplitude of each of the yaw motors relative to each other and relative to at least one threshold based on the current signal from each of the current monitors. The fault detection algorithm further determines a fault condition associated with at least one yaw mechanical drive component of the wind turbine based on the comparison of the current amplitude of each of the yaw motors relative to each other and relative to at least one threshold.

An embodiment of the present invention discloses a warning method and device for a battery. The method includes: obtaining target battery data corresponding to a battery to be detected; for each battery to be detected, using the target battery data corresponding to the battery to be detected and a preset battery fault detection algorithm to determine first fault detection information corresponding to the battery to be detected; using the target battery data corresponding to all batteries to be detected and a target battery data detection model to determine second fault detection information corresponding to each battery to be detected; based on the first fault detection information and the second fault detection information, determining fault verification redundancy information corresponding to each battery to be detected; and outputting the first fault detection information, the second fault detection information, and the fault verification redundancy information to achieve accurate and safe warning of a power battery and improve the safety of battery use.

The invention relates to a vending machine fault early warning method and system based on distributed edge computing, and belongs to the technical field of Internet of Things. The system comprises an equipment management unit, an edge analysis unit, a cloud processing and analysis unit and a visual application unit. The equipment management unit monitors key parameters such as temperature, pressure, liquid level and vibration through the sensor network module, and the edge calculation module operates a lightweight fault detection algorithm to perform local data preprocessing and preliminary diagnosis; the edge analysis unit realizes multi-dimensional data fusion and collaborative decision; the cloud processing and analysis unit performs hierarchical storage, model training and task allocation; and the visual application unit realizes dynamic visualization of the running state of the equipment. Early warning and intelligent management of faults of the automatic soft drink vending machine are achieved, and the equipment operation and maintenance efficiency and the user service experience are improved.

The invention belongs to the technical field of fault identification, and particularly relates to a low-voltage series arc fault identification method based on multi-frequency electrical quantity amplitude-phase combination, which comprises the following steps: each downstream monitoring point monitors the voltage waveform distortion rate of adjacent cyclic waves in real time, and judges whether the voltage waveform distortion rate is greater than or equal to a set distortion rate threshold value; if yes, starting a fault detection algorithm; carrying out asynchronous-in-phase alignment and difference on the voltage data of the downstream monitoring point and the corresponding upstream monitoring point to obtain a differential voltage, carrying out low-pass filtering, and calculating voltage characteristic energy; obtaining a voltage characteristic energy minimum value point, obtaining voltage energy peak values in the left window and the right window, and calculating corresponding phases; judging a peak threshold value and a fault sensitive domain, and counting and accumulating; and if the counting accumulation exceeds the counting threshold value, a fault is determined. According to the invention, the sensitivity, accuracy and anti-interference capability of power grid side fault detection can be improved.

Fault detection system (112) for a drive system (110) with a selective catalytic reduction unit, SCR (206), comprising: a memory (54) configured to store a fault detection algorithm relating to nitrogen oxides, NOx, in the propulsion system (110); and a processor (50) which is operationally coupled to the memory (54) and programmed to: (a) to obtain a first NOx concentration value relating to an upstream input (203) of the SCR unit (206) and (b) a second NOx concentration value relating to a downstream output (205) of the SCR unit (206); (a) and (b) to be temporarily stored in response to receiving (a) and (b) and to repeat the retrieval and buffering until buffered N values ​​have been obtained, where N is a pre-programmed number; to calculate a front-end average of NOx and a front-end standard deviation using the cached N-values; to calculate a downstream average of NOx and a downstream standard deviation using the cached N values; to use the upstream average of NOx, the upstream standard deviation, the downstream average of NOx and the downstream standard deviation as input for the fault detection algorithm relating to nitrogen oxides, NOx, in the propulsion system (110) in order to generate a linear correlation factor; and A non-acceptable part with the best performance, BPU, is detected when the linear correlation factor is greater than a pre-programmed error threshold; where the error detection algorithm includes rules for generating a Pearson correlation; and where the processor (50) is further programmed to: In addition to (a) and (b), to obtain: (c) an upstream mass flow rate value, (d) an upstream temperature value, (e) a downstream mass flow rate value, (f) a downstream temperature value, and (g) a urea value; and Temporarily storing (a), (b), (c), (d), (e), (f) and (g) in response to obtaining (a), (b), (c), (d), (e), (f) and (g); and where each of the cached N values ​​comprises a respective (a), (b), (c), (d), (e), (f) and (g).

One example includes a wind turbine yaw control fault detection system. The system includes current monitors that are each configured to monitor a current amplitude of a respective one of a plurality of yaw motors of a wind turbine and to generate a current signal that is indicative of the respective current amplitude. The system further includes a processor to compare the current amplitude of each of the yaw motors relative to each other and relative to at least one threshold based on the current signal from each of the current monitors. The fault detection algorithm further determines a fault condition associated with at least one yaw mechanical drive component of the wind turbine based on the comparison of the current amplitude of each of the yaw motors relative to each other and relative to at least one threshold.

The application discloses a kind of terminal equipment's fault detection method, system and medium, the method includes the following steps: configuring fault detection program, fault detection equipment and fault detection algorithm;Setting detection reference value, first detection information and second detection information;Detection terminal equipment's start condition, based on fault detection program, fault detection algorithm, detection reference value, first detection information and start condition executes grounding detection step, obtains detection result;Based on fault detection program, fault detection equipment, fault detection algorithm, detection reference value, second detection information and detection result executes equipment operation detection operation or wiring abnormality detection operation;The application can intelligently detect the equipment grounding anomaly of terminal equipment under different operating conditions, and intelligently voice prompts the detection result, finally whether the fault cause of terminal equipment is caused by physical wiring error is judged intelligently, extremely convenient, improve the security of terminal equipment.

The present invention relates to a two-point suspension test bench for maglev trains, comprising an elastic track beam system with unevenness, a sensor-redundant suspension platform, a connecting device, and a counterweight platform. The elastic track beam system can simulate track unevenness and vehicle-track coupling vibration conditions; the suspension platform employs redundant sensor installation to simulate sensor failure; the electromagnet coils on the suspension platform are connected in two parts in series to simulate electromagnet failure; the connecting device connects the two suspension platforms; and the counterweight platform can simulate load-changing conditions by adding or removing counterweights. Compared to existing technologies, the two-point suspension platform designed in the present invention can selectively simulate various actual operating conditions, allowing for research on suspension control or fault detection algorithms, more closely resembling actual conditions. Algorithms designed based on this test bench are more practically applicable.

Embodiments of the present application propose a photovoltaic fault automatic detection method based on multispectral fusion. Hot spot and diode conduction fault are one of the most common faults in photovoltaic modules. Due to dust accumulation, tree shade and foreign matter obstruction, etc. can also cause uneven heat of photovoltaic panel. Directly using infrared image to identify photovoltaic fault has a large error. This brings great threat to the safety and economic operation of photovoltaic power station. This paper proposes a photovoltaic fault detection algorithm based on infrared-visible light fusion. The photovoltaic area of infrared and visible light images is segmented. Then considering the obvious corner features of the photovoltaic panel, the SIFT feature point extraction algorithm is used to register the infrared-visible light image. Then the infrared and visible light images are fused and the features are extracted by using the Transformer multi-head attention mechanism. Then the fused image after feature extraction and enhancement is sent into the deep learning network for defect recognition.