LED large screen fault prediction system and method based on AI
By collecting data from components and analyzing AI algorithms, multi-dimensional data of LED screens can be monitored in real time, solving the problem that manual inspections cannot capture transient faults and identify minor defects in real time, thus achieving efficient fault detection and prediction.
Patent Information
- Application Number
- CN202511261770.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-05
- Publication Date
- 2025-12-16
AI Technical Summary
In existing technologies, fault detection of LED screens relies on manual inspections, which have limited frequency and cannot capture transient faults in real time. Manual identification is not sensitive enough to minor defects, resulting in shortcomings in detection efficiency and blind spots in coverage.
Multi-dimensional data is collected using a 4K high-definition camera, an RS485 communication acquisition device, and a Hall current sensor. Data analysis is performed using an improved U-Net network and an LSTM-Attention hybrid model to generate operation and maintenance decisions and execute hierarchical alarms.
It has improved the ability to capture instantaneous faults in LED screens, accurately identify minute defects, and improve detection efficiency and coverage.
Smart Images
Figure CN121144690A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent display device operation and maintenance, and particularly relates to an AI-based LED large-screen fault prediction system and method. BACKGROUND
[0002] In the technical field of intelligent display device operation and maintenance, commercial LED display screens have been widely used in outdoor advertising, indoor meetings, stage performances and other scenarios due to their high brightness, wide viewing angle, long service life and other advantages. With the expansion of the size of LED large screens and the complication of application scenarios, their operation stability and operation and maintenance efficiency have become the focus of the industry.
[0003] The current mainstream LED large-screen fault detection still mainly relies on traditional manual inspection, which depends on operation and maintenance personnel to regularly check the picture state and check the equipment lines on site.
[0004] However, in the above existing mode, the frequency of manual inspection is limited (usually 1-2 times a day), which cannot capture instantaneous faults in real time, and the sensitivity of manual identification to minor defects is insufficient. For single-point dead lights, local slight color deviation and other hidden faults, it is easy to miss due to differences in personnel experience or visual fatigue. Based on the above, the detection has limitations, and there are obvious efficiency short boards and coverage blind spots. SUMMARY
[0005] The purpose of the present application is to provide an AI-based LED large-screen fault prediction system and method, which aims to solve the technical problems of the prior art, such as limited frequency of manual inspection (usually 1-2 times a day), inability to capture instantaneous faults in real time, insufficient sensitivity of manual identification to minor defects, and easy to miss due to differences in personnel experience or visual fatigue. For single-point dead lights, local slight color deviation and other hidden faults, it is easy to miss due to differences in personnel experience or visual fatigue. Based on the above, the detection has limitations, and there are obvious efficiency short boards and coverage blind spots.
[0006] To achieve the above purpose, an AI-based LED large-screen fault prediction system is adopted, which comprises a collection component, a detection module, an AI decision center and an alarm output module; wherein:
[0007] The collection component is responsible for collecting the original data of the LED screen operation and is the system data input source;
[0008] The detection module is used for targeted analysis and processing of the original data collected by the hardware;
[0009] The AI decision center is used for core operation and decision of the system, receives the output result of the detection module, and generates subsequent processing instructions through comprehensive analysis and judgment by AI algorithm;
[0010] The alarm output module is used for executing specific alarms and operations according to the instructions of the AI decision center.
[0011] The acquisition component comprises a 4K high-definition camera, an RS485 communication collector and a Hall current sensor.
[0012] The 4K high-definition camera is used for collecting LED screen display picture data to provide basic information for picture anomaly detection.
[0013] The RS485 communication collector is used for collecting LED screen control signal data to support signal consistency analysis.
[0014] The Hall current sensor is used for collecting LED screen power supply current data for current ripple monitoring.
[0015] The detection module comprises a picture anomaly detection module, a signal consistency analysis module and a current ripple monitoring module.
[0016] The picture anomaly detection module is used for receiving 4K high-definition camera data to identify picture-level anomalies.
[0017] The signal consistency analysis module is used for receiving RS485 communication collector data to detect the stability and consistency of control signals.
[0018] The current ripple monitoring module is used for receiving Hall current sensor data to monitor whether the current ripple is abnormal.
[0019] The application further provides an AI-based LED large screen fault prediction method, comprising the following steps:
[0020] Collecting multi-dimensional original data in the operation process of the LED large screen:
[0021] Cleaning and standardizing the collected original data to ensure that the data quality meets the AI analysis requirements.
[0022] Analyzing the preprocessed data, performing anomaly identification and health degree evaluation.
[0023] Generating operation and maintenance decisions according to the AI intelligent analysis results and executing them through a hierarchical alarm mechanism.
[0024] Periodically verifying the accuracy and effectiveness of fault prediction and optimizing the algorithm model based on feedback data.
[0025] In the step of collecting multi-dimensional original data in the operation process of the LED large screen:
[0026] The 4K high-definition camera is deployed according to the standard of an installation angle of 30°±5° and an ambient illuminance of not less than 300 lux, real-time collection of visual data of pixel lighting state, color uniformity and picture splicing gap of the LED large screen, and storage of the collected data in the form of H.265 video stream;
[0027] The RS485 communication collector is connected with the LED large screen control box based on the Modbus-RTU protocol, and parameters such as transmission delay, signal amplitude deviation and data frame loss rate of the control signal are collected, with a sampling interval of 100 ms by default, so as to accurately capture the signal integrity of the driving IC and control card components.
[0028] The Hall current sensor is used to independently sample each phase line of the power supply loop of the LED large screen, with a fixed sampling frequency of 1 kHz, so as to collect electrical parameters such as current ripple, peak current and no-load current.
[0029] In the step of cleaning and standardizing the collected raw data to ensure that the data quality meets the requirements of AI analysis:
[0030] For picture data, invalid frames caused by sudden light changes and lens stains are removed; for control signal and current data, an outlier rejection algorithm is used to filter extreme data caused by transient device interference and retain valid operation data.
[0031] The picture data is converted into a pixel point grayscale value matrix, the control signal parameters are normalized to the interval of 0-1, and the current data is standardized according to the actual current / rated current, so as to unify the data dimensions.
[0032] In the step of analyzing the preprocessed data, identifying abnormalities and evaluating health degree:
[0033] The standardized picture data is input into the improved U-Net network, the feature channel weight is dynamically adjusted to enhance the recognition ability of small pixel defects, and the abnormal area coordinates and defect types are output.
[0034] For the standardized control signal data, the sliding window algorithm is used to calculate the mean and variance of the signal transmission delay, and the support vector machine model is used to classify the features of signal amplitude deviation and frame loss rate to identify intermittent failure of driving IC and communication failure of control card.
[0035] The picture abnormality analysis result, signal consistency analysis result and current data calculation result are input into the LSTM-Attention hybrid model.
[0036] In the step of periodically verifying the accuracy and effectiveness of fault prediction and optimizing the algorithm model based on feedback data:
[0037] Statistics monthly fault prediction accuracy, fault response time, unit area maintenance cost, and compares with the traditional scheme to verify the effect;
[0038] Collect fault cases and data in the operation and maintenance process, supplement to the algorithm training data set, optimize the parameters of the improved U-Net network and LSTM-Attention model.
[0039] The AI-based LED large screen fault prediction system and method of the present application adopts the collection component, the detection module, the AI decision center and the alarm output module to perform the following steps: collecting multi-dimensional original data in the running process of the LED large screen: cleaning and standardizing the collected original data to ensure that the data quality meets the AI analysis requirements; analyzing the preprocessed data, identifying abnormalities and assessing health degree; generating operation and maintenance decisions according to the AI intelligent analysis results, and executing through a hierarchical alarm mechanism; verifying the accuracy and effectiveness of fault prediction regularly, and optimizing the algorithm model based on feedback data, through the above-mentioned way, the instantaneous fault capture capability is improved, and the precise identification of small defects is realized, the detection efficiency and the comprehensive optimization of the coverage range are improved. BRIEF DESCRIPTION OF DRAWINGS
[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.
[0041] Figure 1 is the structural principle diagram of the AI-based LED large screen fault prediction system of the present application.
[0042] Figure 2 is the step flowchart of the AI-based LED large screen fault prediction method of the present application.
[0043] Figure 3 is the step flowchart of S500 of the present application.
[0044] Figure 4 is the step flowchart of S600 of the present application.
[0045] Figure 5 is the step flowchart of S700 of the present application.
[0046] Figure 6 is the step flowchart of S900 of the present application.
[0047] 100 - acquisition component, 200 - detection module, 300 - AI decision center, 400 - alarm output module, 101 - 4K high-definition camera, 102 - RS485 communication collector, 103 - Hall current sensor, 201 - picture anomaly detection module, 202 - signal consistency analysis module, 203 - current ripple monitoring module. DETAILED DESCRIPTION
[0048] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, the same numbers are used to indicate the same or similar components. The embodiments described in the following exemplary embodiments are not meant to represent all embodiments consistent with the present application.
[0049] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the description of the application and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
[0050] It should be understood that although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a particular order or hierarchy. These terms are used only to distinguish one from another. For example, a first information can be termed a second information, and similarly, a second information can be termed a first information, without departing from the scope of the present application. Depending on the context, the word "if' as used herein can be interpreted to mean "when" or "in response to determining" or "in response to a determination".
[0051] Referring to Figure 1 The present application also provides an AI-based LED large screen fault prediction system, comprising an acquisition component 100, a detection module 200, an AI decision center 300 and an alarm output module 400; wherein:
[0052] The acquisition component 100 is responsible for collecting original data of LED screen operation, and is a system data input source;
[0053] The detection module 200 is used for targeted analysis and processing of the original data collected by the hardware;
[0054] The AI decision center 300 is used for core operation and decision of the system, receives the output result of the detection module 200, and generates subsequent processing instructions through comprehensive analysis and judgment by AI algorithm;
[0055] The alarm output module 400 is used for executing specific alarm and operation according to the instruction of the AI decision center 300.
[0056] In the embodiment, first, the collection component 100 is responsible for collecting original data of LED screen operation, which is a system data input source; then, the detection module 200 performs targeted analysis and processing on the original data collected by hardware; then, the AI decision center 300 is responsible for core operation and decision of the system, receives the output result of the detection module 200, performs comprehensive analysis and judgment through AI algorithm, and generates subsequent processing instruction; then, the alarm output module 400 executes specific alarm and operation according to the instruction of the AI decision center 300.
[0057] Further, the collection component 100 includes a 4K high-definition camera 101, an RS485 communication collector 102 and a Hall current sensor 103; wherein:
[0058] The 4K high-definition camera 101 is used for collecting LED screen display picture data to provide basic information for picture abnormality detection;
[0059] The RS485 communication collector 102 is used for collecting LED screen control signal data to support signal consistency analysis;
[0060] The Hall current sensor 103 is used for collecting LED screen power supply current data for current ripple monitoring.
[0061] In the embodiment, the 4K high-definition camera 101 collects LED screen display picture data to provide basic information for picture abnormality detection; the RS485 communication collector 102 collects LED screen control signal data to support signal consistency analysis; and the Hall current sensor 103 collects LED screen power supply current data for current ripple monitoring.
[0062] Further, the detection module 200 includes a picture abnormality detection module 201, a signal consistency analysis module 202 and a current ripple monitoring module 203; wherein:
[0063] The picture abnormality detection module 201 is used for receiving 4K high-definition camera 101 data to identify picture level abnormalities;
[0064] The signal consistency analysis module 202 is used for receiving RS485 communication collector 102 data to detect stability and consistency of control signals;
[0065] The current ripple monitoring module 203 is used for receiving Hall current sensor 103 data to monitor whether current ripple is abnormal.
[0066] In the embodiment, the picture anomaly detection module 201 receives 4K high-definition camera 101 data, identifies picture-level anomalies; the signal consistency analysis module 202 receives RS485 communication collector 102 data, detects the stability and consistency of the control signal; and the current ripple monitoring module 203 receives Hall current sensor 103 data, monitors whether there is an anomaly in the current ripple.
[0067] Please refer to Figures 2 to 6 The application also provides an AI-based LED large-screen fault prediction method, which comprises the following steps:
[0068] S500: Collecting multi-dimensional original data in the running process of the LED large screen.
[0069] In the embodiment, multi-dimensional original data in the running process of the LED large screen is collected, and the specific process is as follows:
[0070] S501: A 4K high-definition camera is used to be deployed according to the standard of an installation inclination angle of 30°±5° and an environmental illumination of not less than 300 luxes, real-time collection of visual data of pixel lighting states, color uniformity and picture splicing gaps of the LED large screen, and storage of the collected data in the H.265 video stream format;
[0071] S502: An RS485 communication collector is used to establish a connection with the LED large-screen control box based on the Modbus-RTU protocol, collect parameters such as transmission delay, signal amplitude deviation and data frame loss rate of the control signal, and set the sampling interval to 100 ms by default to accurately capture the signal integrity of the driving IC and control card components;
[0072] S503: A Hall current sensor is used to independently sample each phase line of the power supply circuit of the LED large screen, the sampling frequency is fixed at 1 kHz, and electrical parameters such as current ripple, peak current and no-load current are collected.
[0073] In the above process, a 4K high-definition camera is used to be deployed according to the standard of an installation inclination angle of 30°±5° and an environmental illumination of not less than 300 luxes, real-time collection of visual data of pixel lighting states, color uniformity and picture splicing gaps of the LED large screen, and storage of the collected data in the H.265 video stream format; an RS485 communication collector is used to establish a connection with the LED large-screen control box based on the Modbus-RTU protocol, collect parameters such as transmission delay, signal amplitude deviation and data frame loss rate of the control signal, and set the sampling interval to 100 ms by default to accurately capture the signal integrity of the driving IC and control card components; and a Hall current sensor is used to independently sample each phase line of the power supply circuit of the LED large screen, the sampling frequency is fixed at 1 kHz, and electrical parameters such as current ripple, peak current and no-load current are collected.
[0074] S600: Cleans and standardizes the collected raw data to ensure that the data quality meets the requirements of AI analysis.
[0075] In this implementation, the collected raw data is cleaned and standardized to ensure that the data quality meets the requirements of AI analysis. The specific process is as follows:
[0076] S601: For image data, remove invalid frames caused by sudden changes in light or lens smudges; for control signal and current data, use an outlier removal algorithm to filter out extreme data caused by momentary interference from the device and retain valid operating data.
[0077] S602: Converts image data into a pixel grayscale matrix, normalizes control signal parameters to the 0-1 range, and standardizes current data according to actual current / rated current, unifying data dimensions.
[0078] In the above process, invalid frames caused by sudden changes in light or lens smudges are removed from the image data; for control signals and current data, an outlier elimination algorithm is used to filter out extreme data caused by instantaneous interference from the equipment and retain valid operating data. The image data is converted into a pixel grayscale matrix, the control signal parameters are normalized to the 0-1 range, and the current data is standardized according to the actual current / rated current to unify the data dimensions.
[0079] S700: Analyzes the preprocessed data to identify anomalies and assess health status.
[0080] In this embodiment, the preprocessed data is analyzed to identify anomalies and assess health status. The specific process is as follows:
[0081] S701: Input the standardized image data into the improved U-Net network, and enhance the ability to identify tiny pixel defects by dynamically adjusting the feature channel weights, and output the coordinates of the abnormal area and the defect type.
[0082] S702: For the standardized control signal data, the mean and variance of signal transmission delay are calculated using the sliding window algorithm. The support vector machine model is used to classify the characteristics of signal amplitude deviation and frame loss rate, and to identify intermittent failures of the driver IC and communication failures of the control card.
[0083] S703: Input the results of image anomaly analysis, signal consistency analysis, and current data calculation into the LSTM-Attention hybrid model.
[0084] In the above process, the standardized image data is input into the improved U-Net network. By dynamically adjusting the feature channel weights, the ability to identify tiny pixel defects is enhanced, and the coordinates of abnormal areas and defect types are output. For the standardized control signal data, the mean and variance of signal transmission delay are calculated using the sliding window algorithm. The support vector machine model is used to classify the features of signal amplitude deviation and frame loss rate to identify intermittent failures of the driver IC and communication faults of the control card. The image anomaly analysis results (image integrity, value 0-1), signal consistency analysis results (signal deviation, value 0-1), and current data calculation results (current fluctuation coefficient, value 0-1) are input into the LSTM-Attention hybrid model in the dimension of (time series length, 3). The model captures the temporal dependencies through LSTM and combines the attention mechanism to weight key features, outputting a device health score of 0-100. A score below 60 is judged as a potential fault risk.
[0085] S800: Generates operation and maintenance decisions based on AI intelligent analysis results and executes them through a hierarchical alarm mechanism.
[0086] In this implementation, the AI decision center receives the device health score and the anomaly analysis results of each module, and dynamically optimizes the maintenance cycle by combining additional parameters such as ambient humidity and cumulative running time. The health score is 90-100 (no maintenance required), 70-89 (monthly maintenance), 50-69 (weekly maintenance), and 0-49 (emergency maintenance).
[0087] Tiered alarm execution: Red alarm: When the equipment health score is below 40 points, or serious abnormalities such as the entire large screen going black or current exceeding the standard by more than 30% occur, a red alarm is immediately triggered, sending SMS and telephone alarms to maintenance personnel, and automatically sending a shutdown command to the LED large screen control box to prevent equipment damage; Yellow alarm: When the equipment health score is between 50 and 69 points, or there are minor abnormalities such as 1 to 2 dead LEDs or slightly excessive signal delay, a weekly maintenance plan is generated, and pending tasks are marked on the maintenance management platform to remind maintenance personnel to carry out maintenance work according to the plan.
[0088] S900: Regularly verify the accuracy and effectiveness of fault prediction and optimize the algorithm model based on feedback data.
[0089] In this embodiment, the accuracy and effectiveness of fault prediction are periodically verified, and the algorithm model is optimized based on feedback data. The specific process is as follows:
[0090] S901: Statistically analyze monthly fault prediction accuracy, fault response time, and maintenance cost per unit area, and verify the effectiveness by comparing with traditional solutions;
[0091] S902: Collect fault cases and data during operation and maintenance, supplement them to the algorithm training dataset, and optimize the parameters of the improved U-Net network and LSTM-Attention model.
[0092] In the above process, the monthly fault prediction accuracy, fault response time, and unit area maintenance cost are statistically analyzed and compared with traditional solutions to verify the effectiveness; and fault cases and data in the operation and maintenance process are collected to supplement the algorithm training dataset and optimize the parameters of the improved U-Net network and LSTM-Attention model.
[0093] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.
[0094] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.
Claims
1. An AI-based LED screen fault prediction system, characterized in that, It includes a data acquisition component, a detection module, an AI decision center, and an alarm output module; among which: The acquisition component is responsible for collecting raw data from the operation of the LED screen and serves as the system's data input source. The detection module is used to perform targeted analysis and processing on the raw data collected by the hardware. The AI decision center is used for the core calculations and decisions of the system, receives the output results of the detection module, performs comprehensive analysis and judgment through AI algorithms, and generates subsequent processing instructions. The alarm output module is used to execute specific alarms and operations according to the instructions of the AI decision center.
2. The AI-based LED screen fault prediction system as described in claim 1, characterized in that, The acquisition components include a 4K high-definition camera, an RS485 communication acquisition unit, and a Hall current sensor; wherein: The 4K high-definition camera is used to collect data from the LED screen display, providing basic information for detecting image anomalies. The RS485 communication acquisition device is used to acquire LED screen control signal data and support signal consistency analysis. The Hall current sensor is used to collect LED screen power supply current data for current ripple monitoring.
3. The AI-based LED screen fault prediction system as described in claim 2, characterized in that, The detection module includes a screen anomaly detection module, a signal consistency analysis module, and a current ripple monitoring module; wherein: The image anomaly detection module is used to receive data from a 4K high-definition camera and identify anomalies at the image level. The signal consistency analysis module is used to receive data from the RS485 communication acquisition device and detect the stability and consistency of the control signal. The current ripple monitoring module is used to receive data from the Hall current sensor and monitor whether there are any abnormalities in the current ripple.
4. An AI-based LED screen fault prediction method, applied to the AI-based LED screen fault prediction system as described in claim 3, characterized in that, Includes the following steps: Collect multi-dimensional raw data during the operation of the LED screen: The collected raw data is cleaned and standardized to ensure that the data quality meets the requirements of AI analysis. Analyze the preprocessed data to identify anomalies and assess health status; Operation and maintenance decisions are generated based on AI intelligent analysis results and executed through a tiered alarm mechanism; The accuracy and effectiveness of fault prediction are verified regularly, and the algorithm model is optimized based on feedback data.
5. The AI-based LED screen fault prediction method as described in claim 4, characterized in that, In the process of collecting multi-dimensional raw data during the operation of the LED screen: 4K high-definition cameras are deployed according to the standard of installation tilt angle of 30°±5° and ambient illuminance of not less than 300 lux. The cameras collect visual data in real time on the pixel illumination status, color uniformity and screen splicing gap of the LED screen. The collected data is stored in H.265 video stream format. The RS485 communication acquisition device establishes a connection with the LED screen control box based on the Modbus-RTU protocol, and collects parameters such as transmission delay, signal amplitude deviation, and data frame loss rate of control signals. The sampling interval is set to 100ms by default, accurately capturing the signal integrity of the driver IC and control card components. Hall effect current sensors are used to independently sample each phase of the LED screen power supply circuit, with a fixed sampling frequency of 1 kHz, to collect electrical parameters such as current ripple, peak current, and no-load current.
6. The AI-based LED screen fault prediction method as described in claim 4, characterized in that, In the process of cleaning and standardizing the collected raw data to ensure that the data quality meets the requirements of AI analysis: For image data, invalid frames caused by sudden changes in light or lens smudges are removed; for control signals and current data, an outlier removal algorithm is used to filter out extreme data caused by momentary interference from the equipment and retain valid operating data. The image data is converted into a pixel grayscale matrix, the control signal parameters are normalized to the 0-1 range, and the current data is standardized according to the actual current / rated current to unify the data dimensions.
7. The AI-based LED screen fault prediction method as described in claim 4, characterized in that, In the process of analyzing the preprocessed data, performing anomaly identification and health assessment: The standardized image data is input into the improved U-Net network. By dynamically adjusting the feature channel weights, the ability to identify tiny pixel defects is enhanced, and the coordinates of the abnormal area and the defect type are output. For the standardized control signal data, the mean and variance of signal transmission delay are calculated using the sliding window algorithm. The support vector machine model is used to classify the characteristics of signal amplitude deviation and frame loss rate, and to identify intermittent failures of the driver IC and communication failures of the control card. The results of image anomaly analysis, signal consistency analysis, and current data calculation are input into the LSTM-Attention hybrid model.
8. The AI-based LED screen fault prediction method as described in claim 4, characterized in that, In the steps of periodically verifying the accuracy and effectiveness of fault prediction and optimizing the algorithm model based on feedback data: The monthly fault prediction accuracy, fault response time, and unit area maintenance cost are statistically analyzed and compared with traditional solutions to verify the effectiveness. Collect fault cases and data during operation and maintenance, supplement them to the algorithm training dataset, and optimize the parameters of the improved U-Net network and LSTM-Attention model.