LED driving power supply production quality detection method and device

Abstract

The invention is suitable for the technical field of LED driving power supply manufacturing control, and provides an LED driving power supply production quality detection method and device, and the method is applied to an LED driving power supply production quality detection control system. The system comprises an information acquisition module, a model construction module, a model training module, a model evaluation module, a parameter determination module, a production application module, a wireless communication module, a memory, an alarm, a processing center and an intelligent mobile terminal. The information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, the memory and the alarm are respectively connected with the processing center; the intelligent mobile terminal is in wireless network connection with the wireless communication module in the range of a wireless network or the Internet. The invention further provides an LED driving power supply detection control device. The device can realize automatic production and automatic detection.

Description

Technical Field

[0001] The present invention relates to the field of LED drive power supply production control technology, and in particular to a method and device for detecting the production quality of an LED drive power supply in the field of artificial intelligence. Background Art

[0002] With the application of Internet of Things technology and artificial intelligence technology in many industries, the continuous enrichment of smart home power lines, smart lighting, as an important sub-industry in smart homes, has attracted more and more attention from consumers, and consumers have also put forward new requirements for the intelligence of lighting power supplies. The 5G era and interconnection technology have accelerated the iteration of smart lighting equipment, making consumers have new demands for the intelligence and high-end of LED lighting and the adaptability to various usage scenarios. For LED smart lighting equipment, in addition to power control, various dimming, color adjustment, remote control and interactive functions are also required. In the process of intelligent lighting power supply, lighting technology and the Internet of Things, the Internet, and intelligent software and hardware have achieved cross-border integration, which has promoted the new development of the LED lighting industry. The supporting smart LED lighting driver chip needs to add intelligent system modules such as sensor equipment and remote control devices on the basis of traditional driver chips to meet the new needs of consumers for smart lighting power supplies in the era of smart homes and the Internet of Everything.

[0003] In the production of existing LED driver power supplies, abnormal phenomena such as "aging of electronic components, poor PCB, poor heat dissipation, poor power supply design, lightning damage, grid voltage fluctuation, and solder joint failure" often occur. Due to human design or operating errors or processing errors, the production of LED driver power supplies is often reworked, which makes the production cost high. Therefore, it is necessary to formulate strict production processes for fully automated production to ensure the production quality of each link in the production process, and to prevent quality abnormalities in advance to reduce the number of reworks, so as to ensure the quality of LED driver power supplies and thus reduce production costs. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the purpose of the present invention is to provide a method and device for detecting the production quality of an LED driver power supply. By setting various functional modules such as an information acquisition module, a model construction module, a model training module, a model evaluation module, a parameter determination module, and a production application module, early prevention is carried out from the design by collecting information on abnormal phenomena in daily production quality, presetting production process parameters, and performing automated production and automated detection to reduce human interference and avoid abnormal problems such as "aging of electronic components, poor PCB, poor heat dissipation, poor power supply design, lightning damage, grid voltage fluctuation, and solder joint failure" in the LED driver power supply, thereby improving production efficiency and reducing production costs.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for detecting the production quality of an LED driving power supply is applied to a control system for detecting the production quality of an LED driving power supply. The system comprises an information acquisition module, a model construction module, a model training module, a model evaluation module, a parameter determination module, a production application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal; the information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, the memory, and the alarm are respectively connected to the processing center; the intelligent mobile terminal comprises a smart phone, a tablet computer, and an intelligent remote controller, which are respectively connected to the wireless network of the wireless communication module within the range of a wireless network or the Internet;

[0007] The wireless communication module is provided with a wireless network unit, which is responsible for sending and receiving wireless signals and automatically networking with the intelligent mobile terminal within the effective network range;

[0008] The alarm compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory, and automatically sounds an alarm and notifies to continue training if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to adjust parameters for rework if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to rework or repair if the standard is not met;

[0009] The memory is responsible for storing information of the information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, and the alarm, as well as the model evaluation sub-standard and the LED power supply production quality standard;

[0010] The processing center is responsible for information transmission among the functional modules, alarms, and memories, and is the hub of the system. It compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is passed to the alarm and notified to continue training; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is set as the formal production process parameter; if it does not meet the standard, it is passed to the alarm and notified to adjust the parameters for rework; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is notified that it can be shipped; if it does not meet the standard, it is passed to the alarm and notified to rework or repair;

[0011] The information acquisition module includes a data acquisition unit, a cleaning data unit, an enhancement data unit, an integration data unit, a filtering and denoising unit, a grayscale conversion unit, and a feature extraction unit, which are responsible for acquiring normal and abnormal historical image data of LED driver power production and preprocessing them, and passing them to the model building module;

[0012] The model building module includes a network level unit, a network parameter unit, a hierarchical optimization unit, and a data partitioning unit, which are responsible for determining the model architecture, designing the network level structure, adding auxiliary layers, defining model parameters, and building the model according to the extracted feature information, and passing them to the model training module;

[0013] The model training module includes an input filling unit, a convolution input unit, a pooling conversion unit, a fully connected layer unit, a normalized output unit, an output conversion unit, and a back-propagation unit, which is responsible for training the model according to the collected data and adjusting the model parameters to optimize the model performance to predetermine the power supply production process parameters, and pass them to the model evaluation module;

[0014] The model evaluation module includes a classification labeling unit, a learning training unit, and a model evaluation unit, which is responsible for model verification and evaluation based on the trained model, and adjusting and optimizing the model based on the verification results, and passing it to the parameter determination module;

[0015] The parameter determination module includes a power grid adjustment unit, an output ripple unit, a load adjustment unit, a trial debugging unit, and an abnormality identification unit, which are responsible for performing production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best LED driver power supply production process parameters, and pass them to the production application module;

[0016] The production application module includes a substrate production unit, a mainboard assembly unit, an appearance inspection unit, and a performance inspection unit, which are responsible for producing and inspecting the LED driver power supply according to the set production process parameters and the program instructions of the LED driver power supply production inspection equipment to ensure that the LED driver power supply meets the quality requirements.

[0017] The present invention provides a method for detecting the production quality of an LED driving power supply, comprising the following steps:

[0018] S10, before production, the information acquisition module obtains normal and abnormal historical image data of LED driver power production and pre-processes it for subsequent model construction, and passes it to the model construction module;

[0019] S20, the model building module determines the model architecture, designs the network hierarchy, adds auxiliary layers, and defines model parameters according to the extracted feature information to build the model for subsequent model training and verification, and passes it to the model training module;

[0020] S30, the model training module trains the model according to the collected data and adjusts the model parameters to optimize the model performance to predetermine the power supply production process parameters, ensure the quality stability of the LED driving power supply, and pass it to the model evaluation module;

[0021] S40, the model evaluation module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the model, and passes it to the parameter determination module;

[0022] S50, the parameter determination module performs production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best power supply production process parameters to ensure the subsequent production quality, and passes them to the production application module;

[0023] S60, the production application module performs production and various tests on the LED driver power supply according to the set production process parameters and the LED driver power supply production and testing equipment according to the program instructions to ensure that the LED driver power supply meets the quality requirements.

[0024] Further, the step S10 comprises the following steps:

[0025] S11. The data acquisition unit obtains historical data on circuit performance, mechanical properties, environmental reliability, electromagnetic compatibility, certification standards and production process parameters of similar LED power supplies by networking with the industry database, and transmits it to the cleaning data unit;

[0026] S12, the cleaning data unit cleans the collected historical data to remove abnormal values, duplicate values ​​or missing values ​​in the data to improve the quality and accuracy of the data, and passes it to the enhanced data unit;

[0027] S13, the enhanced data unit increases the quantity and diversity of the LED power supply training data through data enhancement methods such as rotation, scaling, flipping, cropping, and color transformation to improve the generalization ability of the model, and passes it to the integrated data unit;

[0028] S14, the integrated data unit obtains the standardized power data according to the data standardization processing formula "z = (x-μ) / σ, z is the standardized data, x is the original data, μ is the mean of the data, σ is the standard deviation of the data" to improve the stability of the model, and passes it to the conversion data unit;

[0029] S15, the conversion data unit obtains the normalized power data according to the normalization calculation formula "x'=(x-min(x)) / (max(x)-min(x)), x' is the normalized data, x is the original data, min(x) and max(x) are the minimum and maximum values ​​of the data respectively" to improve the model performance, and passes it to the filtering and denoising unit;

[0030] S16, filtering and denoising unit calculates the formula " G(x, y) is a two-dimensional Gaussian function pixel, (x, y) is the pixel coordinate, and σ is the standard deviation. The denoised power image is obtained for grayscale image conversion and passed to the grayscale conversion unit;

[0031] S17, the grayscale conversion unit obtains the grayscale power image according to the grayscale calculation formula "f(i,j) = max(R(i,j), G(i,j), B(i,j)), f(i,j) is the grayscale image, R(i,j), G(i,j), B(i,j) are the original images of three colors respectively", so as to facilitate the subsequent feature extraction, and transmits it to the feature extraction unit;

[0032] S18. The feature extraction unit converts the labeled data into a feature vector useful for model training and extracts the parameter information, appearance structure characteristics, and key features of the protection mark on the label to improve the training speed and performance of the model.

[0033] Further, the step S20 comprises the following steps:

[0034] S21, the network level unit customizes the design and optimizes the network level structure according to the selected convolutional neural network as the model architecture and the specific application scenarios and requirements for subsequent defect detection and quality identification, and passes it to the network parameter unit;

[0035] S22, the network parameter unit adjusts the number of network layers, convolution kernel size, step size, padding method and selects the loss function and optimization algorithm according to the network hierarchy structure to ensure the performance and accuracy of the model, and passes it to the hierarchical optimization unit;

[0036] S23, the hierarchical optimization unit adjusts the size and normalization of the input data by adding a zero padding layer and a normalization layer before and after the convolution layer to improve the model training speed and stability, and passes it to the data partitioning unit;

[0037] S24, the data partitioning unit divides the data set with key feature vectors into a training set, a validation set and a test set according to the extracted data set and establishes a convolutional neural network structure for subsequent model training.

[0038] Further, the step S30 comprises the following steps:

[0039] S31, input filling unit according to the data filling size calculation formula "P H =[(Ho-1)*S H +K H -H i ] / 2,P W=[(Wo-1)*S W +K W -Wi] / 2,P H , P W are the padding sizes in height / width directions, Hi and Wi are the height / width of the input feature map, and K H , K W are the height / width of the convolution kernel, S H , S W are the values ​​of the step length in the height / width direction, Ho and Wo are the height and width of the output feature map respectively. The padding data is obtained and input, and passed to the convolution input unit;

[0040] S32, the convolution input unit obtains the convolution input value and activates the function input according to the convolution input calculation formula "z(t) = ∫x(m)y(tm)dm, z(t) is the output function, x(t) is the input function, y(t) is the convolution kernel function, and dm is the integral differential element of the variable m", and passes it to the pooling conversion unit;

[0041] S33, the pooling conversion unit calculates the maximum pooling formula "Z(i,j)=max(X[i*P S (i+1)*P S ,j*P S (j+1)*P S ]), Z(i,j) is a pixel value in the output feature map after pooling, i and j are the positions in the output feature map, max is the maximum value of all pixel values ​​in the input window, X is the input feature map, P S The maximum pixel value after pooling is obtained by "window size of pooling operation" and enters the fully connected layer to retain important feature information and pass it to the fully connected layer unit;

[0042] S34, the fully connected layer unit obtains the connection layer output data according to the fully connected layer calculation formula "y = f (∑ (wn*xn) + b), y is the output result, f is the activation function, wn is the weight of the nth input feature, xn is the nth input feature, n is the dimension of the input feature, and b is the bias" and enters the output layer for subsequent output calculation and is passed to the normalized output unit;

[0043] S35, the normalized output unit calculates the formula "h = φ (B N (Wx+b)), h is the output of the fully connected layer, φ is the activation function, B N is the batch normalization operator, W is the weight parameter, x is the input of the fully connected layer, and b is the bias parameter. The output of Batch Norm is obtained and passed to the output conversion unit;

[0044] S36, the output conversion unit obtains the output value size after convolution according to the output layer conversion calculation formula "N = (P-F + 2C) / S + 1, N is the output size after convolution, P is the input size before convolution, F is the convolution kernel size, C is the number of layers of 0 added around the image, and S is the step size" for subsequent model analysis and training, and passes it to the back propagation unit;

[0045] S37. The back-propagation unit obtains the loss function value and back-propagates it according to the loss function calculation formula "L = max(0, m + y*(f(x) - b)), L is the loss function value, m is the hyperparameter, y is the sample label (0 or 1), f(x) is the predicted value of the model, and b is the classification boundary" to optimize the performance of the network model.

[0046] Further, the step S40 comprises the following steps:

[0047] S41, the classification and marking unit divides the data sample into corresponding categories according to the quality abnormality categories of the LED driving power supply and marks them for the purpose of machine learning of the subsequent model, and transmits them to the learning and training unit;

[0048] S42, the learning training unit determines the training model according to the gradient descent data, and inputs the data in the training set into the model for simulation training under different environmental conditions to ensure the accuracy of model recognition, and transmits it to the model evaluation unit;

[0049] S43, the model evaluation unit obtains the comprehensive evaluation score F according to the model evaluation score calculation formula "F = 2 (Ac * Re) / (Ac + Re), F is the training model evaluation score, Ac is the precision, and Re is the recall rate", and transmits it to the processing center;

[0050] S44, the processing center compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is transmitted to the alarm and notified to continue training.

[0051] Further, the step S50 comprises the following steps:

[0052] S51, the power grid adjustment unit calculates the power grid adjustment rate according to the formula "Su=|Umax-Umin| / Uo*100%, Su

[0053] is the power grid regulation rate (%), Umax is the maximum output voltage when the power input voltage changes (V), Umin is the minimum output voltage when the power input voltage changes (V), and Uo is the rated output voltage of the power supply (V). The power grid regulation rate is obtained for subsequent power supply testing and passed to the output ripple unit;

[0054] S52, the output ripple unit obtains the power supply output ripple according to the power supply output ripple calculation formula "Vr = Imax / (Co*f), Vr is the power supply output ripple (V), Imax is the maximum output current of the power supply (A), Co is the power supply output capacitance value (F), and f is the power supply switching frequency (Hz)" for subsequent power supply testing and transmits it to the load adjustment unit;

[0055] S53, the load adjustment unit calculates the power load adjustment rate according to the formula "L R =(U A -U B ) / U B *100%, LR is the power supply load regulation rate (%), UA is the output voltage of the power supply when it is not loaded (V), and UB is the output voltage of the power supply when it is fully loaded (V)” to obtain the power supply load regulation rate for subsequent power supply testing and pass it to the trial debugging unit;

[0056] S54, the trial debugging unit conducts trial production and testing of the LED power supply through the power supply production testing equipment according to the predetermined power supply production process parameters to obtain quality information to ensure the quality accuracy of the LED driving power supply, and transmits it to the processing center;

[0057] S55. The processing center compares the power quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is set as the formal production process parameter; if it does not meet the standard, it is transmitted to the alarm and notified to adjust the parameters for rework.

[0058] S56. The abnormality recognition unit matches the real-time LED driver power production detection image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category for subsequent production improvement and parameter adjustment.

[0059] Further, the step S60 comprises the following steps:

[0060] S61, the substrate production unit produces each layer of the circuit board and mounts the electronic components on the circuit board through the automatic circuit board production line and the automatic placement machine according to the control command and the predetermined process parameters, and transmits the result to the mainboard assembly unit;

[0061] S62, the mainboard assembly unit solders the mounted electronic components to the PCB board through wave soldering or reflow soldering according to the control command and the predetermined process parameters, assembles the components with the heat sink and the housing into a power supply, and transmits the power supply to the appearance inspection unit;

[0062] S63, the appearance inspection unit obtains the image or video of the power supply through the set high-definition camera and converts it into recognizable appearance quality information to ensure that the appearance quality of the power supply meets the customer requirements, and transmits it to the performance inspection unit;

[0063] S64. The performance testing unit tests the circuit, mechanical, environmental reliability, electromagnetic compatibility, and transient protection performance of the power supply through the power supply performance testing equipment to ensure that the power supply quality meets the requirements and transmits it to the processing center;

[0064] S65. The processing center compares the quality information of the power supply with the LED power supply production quality standard stored in the memory: if the standard is met, it is notified that the product can be shipped; if the standard is not met, it is transmitted to the alarm and notified to rework or repair.

[0065] An LED driver power supply production quality detection and control system provided by the present invention also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the LED driver power supply production quality detection method described in any one of the above are implemented; the computer-readable storage medium stores a computer program, and when the computer program is executed by the functional modules, the steps of the LED driver power supply production quality detection method described in any one of the above are implemented.

[0066] The present invention also provides an LED driving power supply detection control device, which is implemented by the above-mentioned LED driving power supply production quality detection method.

[0067] The beneficial effects of the present invention compared with the prior art are as follows:

[0068] By setting up various functional modules such as information acquisition module, model construction module, model training module, model evaluation module, parameter determination module, production application module, etc., we collect information on daily production quality abnormalities from the beginning of design to take preventive measures, preset production process parameters, and carry out automated production and automated testing to reduce human interference. We can prevent LED power supplies from having abnormal problems such as "aging of electronic components, poor PCB, poor heat dissipation, poor power supply design, lightning damage, grid voltage fluctuations, and solder joint failure" in production in advance, thereby improving production efficiency and reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or exemplary technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 It is a schematic diagram of the system module of the present invention;

[0071] Figure 2 This is a schematic diagram of an information acquisition module of the present invention;

[0072] Figure 3 It is a schematic diagram of the model construction module of the present invention;

[0073] Figure 4 This is a schematic diagram of a model training module of the present invention;

[0074] Figure 5 It is a schematic diagram of a model evaluation module of the present invention;

[0075] Figure 6 It is a schematic diagram of a parameter determination module of the present invention;

[0076] Figure 7 It is a schematic diagram of the production application module of the present invention;

[0077] Figure 8 It is a schematic diagram of the method flow control program of the present invention;

[0078] Fig. 9 This is a schematic diagram of step S10 in the method flow of the present invention;

[0079] Fig.10 This is a schematic diagram of step S20 in the method flow of the present invention;

[0080] Fig.11 This is a schematic diagram of step S30 in the method flow of the present invention;

[0081] Fig.12 This is a schematic diagram of step S40 in the method flow of the present invention;

[0082] Fig.13 This is a schematic diagram of step S50 in the method flow of the present invention;

[0083] Fig.14 It is a program schematic diagram of step S60 in the method flow of the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0085] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments:

[0086] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. It should be noted that when a module is referred to as "arranged on" another module, it can be directly on another module or indirectly on the other module. When a module is referred to as "connected to" another module, it can be directly connected to another module or indirectly connected to the other module.

[0087] In the description of this application, "multiple" means two or more, unless otherwise clearly defined. "Several" means one or more, unless otherwise clearly defined. "LED driving power supply" in this application is referred to as "LED power supply" or "driving power supply" or "power supply", all of which refer to the same LED driving power supply.

[0088] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0089] See also Figure 1 As shown, the present invention provides a method for detecting the production quality of an LED driving power supply, which is applied to a control system for detecting the production quality of an LED driving power supply, wherein the system comprises an information acquisition module, a model construction module, a model training module, a model evaluation module, a parameter determination module, a production application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal; the information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, the memory, and the alarm are respectively connected to the processing center; the intelligent mobile terminal comprises a smart phone, a tablet computer, and a smart remote controller, which are respectively connected to the wireless communication module wireless network within the range of a wireless network or the Internet.

[0090] The wireless communication module is provided with a wireless network unit, which is responsible for sending and receiving wireless signals and automatically networking with the smart mobile terminal within the effective network range. The wireless signals include MQTT, CoAP, HTTP, REST API, Zigbee, LoRaWAN, NB-IoT, Bluetooth and other IoT signals.

[0091] The alarm compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory, and automatically sounds an alarm and notifies to continue training if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to adjust parameters for rework if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to rework or repair if the standard is not met.

[0092] The memory is responsible for information storage of the information acquisition module, model construction module, model training module, model evaluation module, parameter determination module, production application module, wireless communication module, alarm, as well as storage of the model evaluation sub-standard and the LED power supply production quality standard.

[0093] The processing center is responsible for information transmission among functional modules, alarms, and memories. It is the hub of the system and compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is passed to the alarm and notified to continue training; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is set as the formal production process parameter; if it does not meet the standard, it is passed to the alarm and notified to adjust the parameters for rework; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is notified that it can be shipped; if it does not meet the standard, it is passed to the alarm and notified to rework or repair.

[0094] See also Figure 2 As shown, the information acquisition module includes a data acquisition unit, a cleaning data unit, an enhancement data unit, an integration data unit, a filtering and denoising unit, a grayscale conversion unit, and a feature extraction unit, which are responsible for acquiring normal and abnormal historical image data of LED driver power production and preprocessing them, and passing them to the model building module.

[0095] Furthermore, the data acquisition unit obtains historical data on circuit performance, mechanical properties, environmental reliability, electromagnetic compatibility, certification standards and production process parameters of similar LED power supplies by networking with an industry database, and passes it to a cleaning data unit; the cleaning data unit cleans the collected historical data to remove abnormal values, duplicate values ​​or missing values ​​in the data, and passes it to an enhancement data unit; the enhancement data unit increases the quantity and diversity of power supply training data by data enhancement methods such as rotation, scaling, flipping, cropping and color transformation, and passes it to an integrated data unit; the integrated data unit obtains standardized power supply data according to the data standardization processing formula "z = (x-μ) / σ", and passes it to a conversion data unit; the conversion data unit obtains normalized power supply data according to the normalization calculation formula "x' = (x-min(x)) / (max(x)-min(x))", and passes it to a filtering and denoising unit; the filtering and denoising unit calculates the power supply data according to the Gaussian filtering method formula The denoised power image is obtained and passed to a grayscale conversion unit; the grayscale conversion unit obtains the grayscale power image according to the grayscale calculation formula "f(i,j)=max(R(i,j),G(i,j),B(i,j))" and passes it to a feature extraction unit; the feature extraction unit converts the labeled data into a feature vector useful for model training and extracts the parameter information, appearance structure characteristics, and key features of the protection mark on the label to improve the training speed and performance of the model.

[0096] See also Figure 3 As shown, the model construction module includes a network hierarchy unit, a network parameter unit, a hierarchical optimization unit, and a data partitioning unit, which is responsible for determining the model architecture, designing the network hierarchy structure, adding auxiliary layers, defining model parameters, and constructing the model based on the extracted feature information, and passing it to the model training module.

[0097] Furthermore, the network hierarchy unit customizes the design and optimizes the network hierarchy structure according to the selected convolutional neural network as the model architecture and according to the specific application scenarios and requirements, and passes it to the network parameter unit; the network parameter unit adjusts the number of network layers, convolution kernel size, step size, padding method and selects the loss function and optimization algorithm according to the network hierarchy structure, and passes it to the hierarchy optimization unit; the hierarchy optimization unit adjusts the size and normalization of the input data by adding a zero padding layer and a normalization layer before and after the convolution layer, respectively, and passes it to the data partitioning unit; the data partitioning unit divides the data set with key feature vectors after extraction into a training set, a validation set and a test set and establishes a convolutional neural network structure for subsequent model training.

[0098] See also Figure 4As shown, the model training module includes an input filling unit, a convolution input unit, a pooling conversion unit, a fully connected layer unit, a normalized output unit, an output conversion unit, and a back-propagation unit, which is responsible for training the model according to the collected data and adjusting the model parameters to optimize the model performance to predetermine the power supply production process parameters and pass them to the model evaluation module.

[0099] Further, the input filling unit obtains the filling data and inputs it according to the data filling size calculation formula "PH = [(Ho-1)*SH + KH-Hi] / 2, PW = [(Wo-1)*SW + KW-Wi] / 2", and passes it to the convolution input unit; the convolution input unit obtains the convolution input value and activates the function input according to the convolution input calculation formula "z(t) = ∫x(m)y(tm)dm", and passes it to the pooling conversion unit; the pooling conversion unit obtains the maximum pixel value after pooling according to the maximum pooling calculation formula "Z(i,j) = max(X[i*PS(i+1)*PS,j*PS(j+1)*PS])" and enters the fully connected layer, and passes it to the fully connected layer unit; the fully connected layer unit obtains the connection layer output data according to the fully connected layer calculation formula "y = f(∑(wn*xn)+b)" and enters the output layer, and passes it to the normalized output unit; the normalized output unit obtains the connection layer output data according to the Batch The Norm layer calculation formula "h = φ (BN (Wx + b))" obtains the output of Batch Norm and passes it to the output conversion unit; the output conversion unit obtains the output value size after convolution according to the output layer conversion calculation formula "N = (P-F + 2C) / S + 1" and passes it to the back propagation unit; the back propagation unit calculates the loss function according to the formula "L = max (0, m + y *

[0100] (f(x)-b))” obtains the loss function value and back-propagates to optimize the performance of the network model.

[0101] See also Figure 5 As shown, the model evaluation module includes a classification labeling unit, a learning training unit, and a model evaluation unit, which is responsible for model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results and passes them to the parameter determination module.

[0102] Furthermore, the classification and marking unit divides the data samples into corresponding categories according to the quality anomaly categories of the LED driving power supply and marks them for subsequent machine learning of the model, and transmits them to the learning and training unit; the learning and training unit determines the training model according to the gradient descent data, and inputs the data in the training set into the model for simulation training under different environmental conditions, and transmits it to the model evaluation unit; the model evaluation unit obtains the F comprehensive evaluation score according to the model evaluation score calculation formula "F=2(Ac*Re) / (Ac+Re)", and transmits it to the processing center.

[0103] See also Figure 6 As shown, the parameter determination module includes a power grid adjustment unit, an output ripple unit, a load adjustment unit, a trial debugging unit, and an abnormality identification unit, which is responsible for conducting production tests under different conditions based on the production process parameters predicted by the trained model to obtain the optimal LED driver power supply production process parameters and pass them to the production application module.

[0104] Further, the power grid adjustment unit obtains the power grid adjustment rate according to the power grid adjustment rate calculation formula "Su=|Umax-Umin| / Uo*100%", and transmits it to the output ripple unit; the output ripple unit obtains the power output ripple according to the power output ripple calculation formula "Vr=Imax / (Co*f)", and transmits it to the load adjustment unit; the load adjustment unit obtains the power load adjustment rate according to the power load adjustment rate calculation formula "LR=(UA-UB) / UB*100%", and transmits it to the trial debugging unit; the trial debugging unit trial-produces and detects the LED driver power supply through the power production detection equipment according to the predetermined power production process parameters to obtain quality information, and transmits it to the processing center; the abnormality recognition unit matches the real-time LED driver power production detection image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category for subsequent production improvement and parameter adjustment.

[0105] See also Figure 7 As shown, the production application module includes a substrate production unit, a mainboard assembly unit, an appearance inspection unit, and a performance inspection unit, which are responsible for producing and inspecting the LED driver power supply according to the program instructions of the LED driver power supply production inspection equipment according to the set production process parameters to ensure that the LED driver power supply meets the quality requirements.

[0106] Furthermore, the substrate production unit produces each layer of the circuit board and mounts the electronic components on the circuit board through the circuit board automatic production line and the automatic placement machine according to the control command and the predetermined process parameters, and transmits them to the mainboard assembly unit; the mainboard assembly unit solders the mounted electronic components to the PCB board through wave soldering or reflow soldering according to the control command and the predetermined process parameters, and assembles them into a power supply with the heat sink and the housing, and transmits them to the appearance inspection unit; the appearance inspection unit obtains the image or video of the power supply through the set high-definition camera and converts it into recognizable appearance quality information, and transmits it to the performance inspection unit; the performance inspection unit respectively detects the circuit, mechanical, environmental reliability, electromagnetic compatibility, and transient protection performance of the power supply through the power supply performance testing equipment, and transmits it to the processing center.

[0107] System operation working principle:

[0108] Before production, the information acquisition module obtains normal and abnormal historical image data of LED driver power supply production and performs preprocessing for subsequent model construction, and passes it to the model construction module; then the model construction module determines the model architecture, designs the network hierarchy, adds auxiliary layers, and defines model parameters according to the extracted feature information to build the model for subsequent model training and verification, and passes it to the model training module; the model is trained according to the collected data through the model training module and the model parameters are adjusted to optimize the model performance to predetermine the power supply production process parameters to ensure the quality stability of the LED driver power supply, and pass it to the model evaluation module; the model is verified and evaluated according to the trained model through the model evaluation module, and the model is adjusted and optimized according to the verification results to improve the performance and accuracy of the model, and passed to the parameter determination module; then the parameter determination module performs production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best power supply production process parameters to ensure the subsequent production quality, and passes it to the production application module; the LED driver power supply production and various tests are performed on the LED driver power supply production and testing equipment according to the set production process parameters according to the program instructions through the production application module to ensure that the LED driver power supply meets the quality requirements.

[0109] When operators or managers are outdoors or in other places, they can use smart mobile terminals to automatically connect to the wireless network or the Internet through the wireless communication module, realize the combination of smart mobile terminals, the Internet and Internet of Things technologies, and use the APP software on the smart mobile terminal or the remote control software on the computer to control the power supply to produce under normal conditions at close range or remotely to meet the user's quality requirements. Managers can remotely control and monitor the power supply production and detection, view the daily operating status, data changes, equipment activity logs, etc., and realize remote connection and control of power supply production by prefabricated deployment codes or adding deployment codes. Remote power supply production detection can be viewed in real time through smart mobile terminals or computers, which improves production efficiency and reduces production costs.

[0110] See also Figure 8 As shown, the present invention provides a method for detecting the production quality of an LED driving power supply, comprising the following steps:

[0111] S10, before production, the information acquisition module obtains normal and abnormal historical image data of LED driver power production and pre-processes it for subsequent model construction, and passes it to the model construction module;

[0112] See also Fig. 9 As shown, the step S10 comprises the following steps:

[0113] S11. The data acquisition unit obtains historical data on circuit performance, mechanical properties, environmental reliability, electromagnetic compatibility, certification standards and production process parameters of similar LED driver power supplies by networking with the industry database, and transmits the data to the cleaning data unit;

[0114] It is further explained that the production process parameters of various LED driver power supplies, as well as the quality information of the corresponding power supply images or videos and the data of the use effects of different users are collected from domestic and foreign LED driver power supply databases through a data collector for classification and collection; the LED driver power supply quality data includes but is not limited to the historical data of different users' evaluation of the use effects of different such LED driver power supplies, and also includes the normal and abnormal quality information of similar LED driver power supplies and the historical data of the use effects of different users obtained from the Internet, social media, professional testing institutions, processing trade platforms, or third-party sharing platforms such as public data sets through web crawlers, sensor collection, manual annotation, data set purchase, crowdsourcing, etc. Acquire high-quality, diverse and rich historical data of similar LED driver power supplies, specifically including historical data of quality information such as raw materials, performance parameters, process parameters, test data and images of LED driver power supplies; the raw material data include shape drawings and processing drawings and data of electronic components of various LED driver power supplies; the performance parameters include voltage, power factor, current and power, work efficiency, protection function, working temperature, waterproof grade, safety structure and other performance data; the process parameters include manufacturing process flow, production equipment control parameters, test equipment control parameters, working temperature, working voltage and other data; the test data include circuit performance, mechanical properties, environmental reliability, electromagnetic compatibility, certification standards and other data.

[0115] S12, the cleaning data unit cleans the collected historical data to remove abnormal values, duplicate values ​​or missing values ​​in the data to improve the quality and accuracy of the data, and passes it to the enhanced data unit;

[0116] It is further explained that due to sensor failure, recording errors or system anomalies in the production and detection of LED driver power supplies, outliers occur, which affect the accuracy of subsequent analysis; missing value processing is performed on data with missing parts; each feature in the data set is detected by writing code to determine the missing values, and the missing value pattern is identified. According to the number and impact of the missing values, the code is written to select to fill or delete samples or variables with missing values, and forward filling or backward filling is used to fill the missing values ​​for time series data; the processing effect is verified by comparing indicators such as data quality before and after processing, accuracy and reliability of the model. If the processing effect is not good, the processing method is reselected or the data is re-cleaned until the best effect is achieved, so as to remove duplicate, erroneous, and invalid data to effectively process missing values ​​for subsequent model construction and analysis.

[0117] S13, the enhanced data unit increases the quantity and diversity of the LED driver power training data through data enhancement methods such as rotation, scaling, flipping, cropping, and color transformation to improve the generalization ability of the model, and passes it to the integrated data unit;

[0118] It is further explained that after the cleaned LED driver power supply image data is rotated by a certain angle through the LED driver power supply image data rotation, multiple pixels will correspond to the same pixel after rotation, resulting in pixel loss in the rotated LED driver power supply image data. Therefore, a mapping is constructed by reverse thinking to map the pixel coordinates after rotation to the pixel coordinates of the original LED driver power supply image data, so that each pixel after rotation can be guaranteed to have a pixel value; each original pixel in the rotated LED driver power supply data is copied and mapped intact to the corresponding four pixels after expansion using a neighborhood interpolation algorithm, thereby retaining all information of the original LED driver power supply image data; the LED driver power supply image data with abnormal orientation after scaling is flipped 180 degrees around the central axis or symmetry axis of the original image to ensure the orientation consistency of the LED driver power supply data; the flipped LED driver power supply image data is cropped to ensure image integrity and clarity.

[0119] S14, the integrated data unit obtains the standardized power data according to the data standardization processing formula "z = (x-μ) / σ, z is the standardized data, x is the original data, μ is the mean of the data, σ is the standard deviation of the data" to improve the stability of the model, and passes it to the conversion data unit;

[0120] It is further explained that the collected raw data of the LED power supply is calibrated, including time synchronization, unit conversion, range adjustment, etc., such as ensuring that all timestamps are in a unified format and synchronized with the system clock; converting data collected by different sensors, detection equipment, etc. to the same unit to ensure data accuracy; using the written code to use the above data standardization processing formula "z = (x-μ) / σ" to convert data from different sources such as dates, numbers, texts, etc. into a standard normal distribution with a mean of 0 and a standard deviation of 1, subtracting the mean and dividing by the standard deviation to make it a unified standard format to ensure data consistency and comparability for subsequent processing; encoding unstructured data and converting it into structured data, or converting text data into numerical representation for model processing and analysis, thereby processing the raw data into a form that meets specific standards for subsequent model training and analysis.

[0121] S15, the conversion data unit obtains the normalized power data according to the normalization calculation formula "x'=(x-min(x)) / (max(x)-min(x)), x' is the normalized data, x is the original data, min(x) and max(x) are the minimum and maximum values ​​of the data respectively" to improve the model performance, and passes it to the filtering and denoising unit;

[0122] It is further explained that the above normalization calculation formula "x'=(x-min(x)) / (max(x)-min(x))" is used by the written code to proportionally map and scale the cleaned data to the specified range of [0,1], convert the data into a unified format or range of data normalization, convert continuous features into discrete features of data discretization, etc. to convert the data to eliminate the total amount difference between different samples or features and the dimensional influence between the data, make the distribution of the data consistent, and confirm whether the data has been normalized to the specified range or distribution as expected, and verify the normalization effect through statistical descriptions such as maximum value and minimum value; the data includes multiple types such as text, pictures, audio, video, etc., to eliminate the dimensional differences between different data, thereby improving the accuracy and efficiency of data analysis, so as to better adapt to subsequent analysis and processing.

[0123] S16, filtering and denoising unit calculates the formula " G(x,y) is the two-dimensional Gaussian function pixel, (x,y) is the pixel coordinate, and σ is the standard deviation” to obtain the denoised power image for grayscale image conversion and pass it to the grayscale conversion unit;

[0124] It is further explained that the Gaussian filter output pixel value obtained according to the Gaussian filter calculation formula is the weighted average of the input pixel values ​​in its neighborhood, and the weight is given by the Gaussian function, where the standard deviation σ determines the width of the Gaussian function, and the Gaussian function is discretized, that is, the weight is calculated within a certain window size, and then these weights are applied to the corresponding neighborhood pixel values ​​of the input image to obtain the output pixel value, and the pixels are arranged from small to large according to the grayscale value and the median is taken as the new grayscale value; the sampling value in the input signal is checked to determine whether it represents the signal itself, and the values ​​in the window are sorted by using an observation window composed of an odd number of samples, the middle value is taken as the output, the earliest value is discarded, a new sample is obtained, and the above calculation process is repeated to reduce the random noise in the image and make the image smoother.

[0125] S17, the grayscale conversion unit obtains the grayscale power image according to the grayscale calculation formula "f(i,j) = max(R(i,j), G(i,j), B(i,j)), f(i,j) is the grayscale image, R(i,j), G(i,j), B(i,j) are the original images of three colors respectively", so as to facilitate the subsequent feature extraction, and transmits it to the feature extraction unit;

[0126] Further explanation: the grayscale calculation formula is used to obtain the grayscale LED driver power image, the sampling value in the input signal is checked to determine whether it represents the signal itself, and the values ​​in the window are sorted by using an observation window composed of an odd number of samples, the middle value is taken as the output, the earliest value is discarded, and a new sample is obtained. The above calculation process is repeated to remove the noise in the LED driver power image or other signals, so as to subsequently extract the key features of the LED driver power image, including power type, performance parameters, waterproof level, appearance structure, safety structure and other information, as well as appearance quality information such as shell integrity, connection port status, identification information, etc.; then the standardized data is classified through the written code, and Automatically label classified data, add accurate labels or annotations to the data, and use them as target variables or features in subsequent model training, and as training sets and validation sets, to ensure that the model has accurate data references during the training process, so that it can learn how to generate corresponding labels based on data features to improve labeling efficiency and accuracy; for some data, users are required to enter the system remotely through their mobile phones to manually add new labels or modify existing labels to ensure that they accurately reflect the essential characteristics of the data, and deploy the labeled data to the corresponding system platform for subsequent application or analysis to improve the accuracy and reliability of the labels, so as to facilitate the extraction of key features in subsequent data, which is more conducive to model training, testing and verification.

[0127] S18. The feature extraction unit converts the labeled data into a feature vector useful for model training and extracts the parameter information, appearance structure characteristics, and key features of the protection mark on the label to improve the training speed and performance of the model.

[0128] It is further explained that the most representative features and the most useful features for the model that can reflect the essential content of the data and have the degree of discrimination and description are selected from the preprocessed power supply data to reduce the dimension of the features, reduce the computational complexity, and improve the performance and generalization ability of the model; after feature extraction, principal component analysis is used to perform feature dimensionality reduction to obtain a large amount of feature information in order to reduce the dimension of the features and reduce the computational complexity; after feature dimensionality reduction, the features are binary encoded to improve the interpretability of the features and the performance of the model; since some information of the original data may be lost after feature extraction and dimensionality reduction, feature reconstruction such as principal component reconstruction or least squares reconstruction is performed to restore the data information as much as possible; evaluation is performed through cross-validation, and the feature extraction method and parameters are adjusted according to the evaluation results to evaluate the quality and effect of the extracted features, improve the training speed and performance of the model, and improve the efficiency of data analysis by reducing the amount or complexity of the data while keeping the original appearance of the data as much as possible.

[0129] S20, the model building module determines the model architecture, designs the network hierarchy, adds auxiliary layers, and defines model parameters according to the extracted feature information to build the model for subsequent model training and verification, and passes it to the model training module;

[0130] See also Fig.10 As shown, the step S20 comprises the following steps:

[0131] S21, the network level unit customizes the design and optimizes the network level structure according to the selected convolutional neural network as the model architecture and the specific application scenarios and requirements for subsequent defect detection and quality identification, and passes it to the network parameter unit;

[0132] Further explanation: Since the convolutional neural network (CNN) is a deep neural network, it is particularly suitable for processing data with a grid topology. In the production of LED driver power supplies, it can automatically learn and extract useful features such as edges, lines, corners, and more complex combined features from images for subsequent defect detection, quality classification, etc.; since the basic structure of CNN includes input layer, convolution layer, activation function layer, pooling layer, fully connected layer, and output layer, these layers can work together in the CNN model of LED driver power supply production to extract and process image features, including: the input layer can receive LE The image data of the LED driver is taken as input; multiple convolution layers are set, each layer contains multiple convolution kernels, which can extract local features in the image. As the network structure deepens, the appearance gradually becomes abstract; adding a ReLU activation function layer after the convolution layer can introduce nonlinear factors and enhance the expression ability of the model; adding a pooling layer after the convolution layer can reduce the dimension of the feature map and reduce the amount of calculation while maintaining the spatial invariance of important features; adding a fully connected layer at the end of the network is used to convert the feature map output by the convolution layer and the pooling layer into a classification result, that is, the category or quality grade of the LED driver.

[0133] S22, the network parameter unit adjusts the number of network layers, convolution kernel size, step size, padding method and selects the loss function and optimization algorithm according to the network hierarchy structure to ensure the performance and accuracy of the model, and passes it to the hierarchical optimization unit;

[0134] It is further explained that the model parameters are defined according to the network layer design, including the size of the convolution kernel, step size, padding method, activation function, etc., as well as the pooling method of the pooling layer, the pooling window size, etc., the number of neurons in the fully connected layer and other layer parameters, which directly affect the performance and accuracy of the model; then the hyperparameters related to the model training, such as the size of the input image, the number of label types, the total training cycle, and the batch size, are set, and adjusted according to the specific data set and task requirements; the optimization algorithms such as SGD and Adam are selected, and the corresponding learning rate, momentum and other parameters are set, which directly affect the training effect and convergence speed of the model to ensure that the model can accurately extract features from the image and perform effective classification or prediction.

[0135] S23, the hierarchical optimization unit adjusts the size and normalization of the input data by adding a zero padding layer and a normalization layer before and after the convolution layer, respectively, to improve the model training speed and stability, and passes it to the data partitioning unit;

[0136] It is further explained that by setting the zero padding layer to pad the edges of the input image with zero values ​​before the convolution operation, the size of the input data is adjusted to facilitate the convolution operation and maintain the image edge information, ensuring that the convolution kernel can be correctly applied to the image boundary, while helping to maintain the image edge information and avoid information loss during the convolution process; zero padding can more effectively process input images of different sizes without complex preprocessing, which helps to control the size of the convolution layer output feature map, making the network design more flexible and controllable, ensuring that the network can adapt to changes in LED driver power supplies of different sizes and resolutions, and accurately extract and process image features; the distribution of the input data is normalized to improve the training speed and stability of the model and accelerate the training process of the neural network. It improves the convergence speed and enhances the stability of the model. The input of the activation function is normalized before the activation function of each layer of the network. The mean and variance of the batch are calculated for each small batch of data, and then the linear calculation results are batch normalized, that is, the mean is subtracted and divided by the standard deviation to ensure that the calculation results conform to the standard normal distribution with a mean of 0 and a variance of 1. Then, translation and scaling operations are performed to adapt to different data distribution requirements, so that the input of the middle layer of the network remains relatively stable, which helps to solve the problem of gradient disappearance or gradient explosion during training, thereby accelerating training and enhancing the stability of the model. It can improve the robustness of the model to the weight initialization method and alleviate the troubles caused by weight initialization selection, so as to improve the generalization ability and training efficiency of the model.

[0137] S24, the data partitioning unit divides the data set with key feature vectors into a training set, a validation set and a test set according to the extracted data set and establishes a convolutional neural network structure for subsequent model training.

[0138] It is further explained that the ratio of the training set, the validation set, and the test set is determined according to the pre-selected segmentation strategy, and the data distribution between the subsets is ensured to be as consistent as possible to avoid introducing bias; in some application scenarios, since the data has time series characteristics, the data set is divided into a training set for training the model, a validation set for adjusting model parameters and selecting the best model, and a test set for evaluating the final performance of the model according to the written code, so as to ensure the consistency of the training and test sets in time; the data set is used in a 9:1 relationship through the deep learning of the convolutional neural network to establish the training set and test set required for the training of the LED driver power quality abnormality recognition pattern prediction model under different environments, thereby establishing a deep convolutional neural network structure: 3 convolutional layers, 3 maximum pooling layers, 2 fully connected layers, 1 softmax layer, 1 output layer, and zero padding layers and normalization layers are added when necessary; since the convolution kernel in the convolutional layer contains weight coefficients, while the pooling layer does not contain weight coefficients, the pooling layer is not an independent layer, the convolutional layer is used to extract local features of the image, the pooling layer is used to reduce the dimension of the feature map, and the fully connected layer is used to integrate features and classify.

[0139] S30, the model training module trains the model according to the collected data and adjusts the model parameters to optimize the model performance to predetermine the power supply production process parameters, ensure the quality stability of the LED driving power supply, and pass it to the model evaluation module;

[0140] See also Fig.11 As shown, the step S30 comprises the following steps:

[0141] S31, input filling unit according to data filling size calculation formula

[0142] "PH = [(Ho-1)*SH + KH-Hi] / 2, PW = [(Wo-1)*SW + KW-Wi] / 2, PH and PW are the padding sizes in the height / width direction, Hi and Wi are the height / width of the input feature map, KH and KW are the height / width of the convolution kernel, SH and SW are the values ​​of the step length in the height / width direction, Ho and Wo are the height and width of the output feature map" to obtain the padding data and input it, and pass it to the convolution input unit;

[0143] It is further explained that after the image is standardized, the LED driving power data first enters the input layer and obtains the padding data through the padding size calculation formula to enter the zero padding layer, fills the edge of the convolution layer input data with zero values, and adjusts the size of the input data for convolution operation to maintain the same spatial dimension of the input and output, thereby preventing the loss of image edge information, so that the spatial dimension of the input data can be kept unchanged during subsequent convolution operations to maintain the image edge information and the processing of subsequent layers; the amount of padding is determined by the size of the convolution kernel, the step size and the padding size. In order to keep the height and width of the input and output images the same, or to reduce the size quickly in continuous convolution operations, the amount is usually set to the convolution kernel size minus the convolution kernel size; the height Hi / width Wi of the input feature map, the height KH / width KW of the convolution kernel, the step value SH / SW in the height / width direction, and the height Ho / width Wo of the output feature map are all obtained through model design parameters.

[0144] S32, the convolution input unit obtains the convolution input value and activates the function input according to the convolution input calculation formula "z(t) = ∫x(m)y(tm)dm, z(t) is the output function, x(t) is the input function, y(t) is the convolution kernel function, and dm is the integral differential element of the variable m", and passes it to the pooling conversion unit;

[0145] It is further explained that the "width" or "increment" of the integral differential element dm when integrating each infinitesimal interval of the variable m, and the cumulative effect on the entire definition domain, namely z(t), is obtained by accumulating the function values ​​on these infinitesimal intervals (namely x(m)y(tm)), and dm is the integral differential element of the variable m, that is, the infinitesimal quantity in the integration process; and the convolution calculation is performed to obtain the convolution value to enter the convolution layer and obtain the activation function value through the ReLU function calculation formula "f(x)=max(0,x+Y)" and enter the activation function layer, where f(x) is the activated function, x is the eigenvalue of the output of the convolution layer or the fully connected layer, and Y is a random variable. When the input x is greater than 0, x is output, when the input x is less than or equal to 0, 0 is output, and when x=0, it is not differentiable, that is, when the input is a positive number, the original value is maintained, and when the input is a negative number, 0 is output; but the ReLU activation function is used after the convolution layer to increase the nonlinearity of the network and promote the model to learn complex patterns.

[0146] S33, the pooling conversion unit obtains the maximum pixel value after pooling according to the maximum pooling calculation formula "Z(i,j)=max(X[i*PS(i+1)*PS,j*PS(j+1)*PS]), Z(i,j) is a pixel value in the output feature map after pooling, i and j are the positions in the output feature map, max is the maximum value of all pixel values ​​in the input window, X is the input feature map, PS is the window size of the pooling operation", enters the fully connected layer to retain important feature information, and passes it to the fully connected layer unit;

[0147] It is further explained that the average value of all values ​​in each area is obtained as output through the maximum pooling calculation formula, which can retain more information in the feature map, help improve the accuracy of the model, retain more detail information, and be used in the deeper layers of the network to ensure the integrity of the information; before maximum pooling, the maximum pixel value after average pooling is obtained according to the average pooling calculation formula "Z(i,j)=mean(X[i*PS(i+1)*PS,j*PS(j+1)*PS])", Z(i,j) is a pixel value in the output feature map after pooling, mean is the average value of all pixel values ​​in the input window, and the maximum value is selected from each area of ​​the input feature map as the output according to the formula, retaining the most significant features in the feature map, reducing the size of the feature map, and losing some useful information. Since the maximum value of each area is focused on, the grain size feature is retained, and features with better classification recognition are selected.

[0148] S34, the fully connected layer unit obtains the connection layer output data according to the fully connected layer calculation formula "y = f (∑ (wn*xn) + b), y is the output result, f is the activation function, wn is the weight of the nth input feature, xn is the nth input feature, n is the dimension of the input feature, and b is the bias" and enters the output layer for subsequent output calculation and is passed to the normalized output unit;

[0149] Further explanation: after pooling, the data is subjected to a full connection layer and the output value after convolution is obtained according to the full connection layer calculation formula "y = f(∑(wn*xn)+b)", where y is the output result, f is the activation function, wn is the weight of the nth input feature, xn is the nth input feature, n is the dimension of the input feature, and b is the bias; the "y = f(∑(w n *x n )+b)” is calculated from “y=f(w*x+b)” to “y=f(w 1 *x 1 +w 2 *x 2 +…+w n *x n +b)” and transformed, where w is the weight, x nThe features extracted by the convolution layer and the pooling layer are integrated and classified through the fully connected layer, so that each neuron is connected to all the neurons in the previous layer, and the output is generated through weighted summation and nonlinear activation function.

[0150] S35, the normalized output unit calculates the formula "h = φ (B N (Wx+b)), h is the output of the fully connected layer, φ is the activation function, B N is the batch normalization operator, W is the weight parameter, x is the input of the fully connected layer, and b is the bias parameter. The output of Batch Norm is obtained and passed to the output conversion unit;

[0151] Further explanation, the "h=φ(B N (Wx+b))” through “B N (x) = γ⊙[(x-μ B ) / σ B ]+β”, where μ B is the mean, σ B is the standard deviation, γ is the stretch parameter, β is the offset parameter, (x-μ B ) / σ B To standardize the normal distribution; the Batch Norm layer (referred to as "BN layer") is placed between the affine transformation and the activation function in the fully connected layer, and the output value of each layer in the network is standardized to make it more obedient to the normal distribution, which can accelerate the training speed of the neural network and help prevent gradient disappearance and gradient explosion; the BN layer reduces the correlation between input data by standardizing the data of each mini-batch, so that the mean of each sample is 0 and the variance is 1, thereby reducing the problem of internal covariate shift, helping to accelerate the training process, improve the stability and generalization ability of the model, and avoid reducing the problem of reducing the representation ability of the neural network.

[0152] S36, the output conversion unit obtains the output value size after convolution according to the output layer conversion calculation formula "N = (P-F + 2C) / S + 1, N is the output size after convolution, P is the input size before convolution, F is the convolution kernel size, C is the number of layers of 0 added around the image, and S is the step size" for subsequent model analysis and training, and passes it to the back propagation unit;

[0153] Further explanation: the output size N is the size of the feature map after the convolution operation, the input size P is the width or height of the input image or feature map, and the padding size PH in the height direction and the padding size P in the width direction are calculated by the above step S21. WThe convolution kernel size F is obtained by calculating the above step S22, the number of layers C of 0 added around the image is the number of layers of 0 added around the input image or feature map to control the size of the output feature map, and is also obtained by the above network construction, and the step size S is the distance that the convolution kernel slides on the input image or feature map obtained by the above calculation.

[0154] S37. The back-propagation unit obtains the loss function value and back-propagates it according to the loss function calculation formula "L = max(0, m + y*(f(x) - b)), L is the loss function value, m is the hyperparameter, y is the sample label (0 or 1), f(x) is the predicted value of the model, and b is the classification boundary" to optimize the performance of the network model.

[0155] To further illustrate, in the loss function calculation formula "L = max(0,m+y*(f(x)-b))", when f(x)≤b, then L = m+y(f(x)-b), and when f(x)>b, then L = 0; in machine learning or optimization problems, f(x) is an optimization target, b is a threshold, and m and y are parameters; when the target value f(x) is less than or equal to the threshold b, then the value of L is equal to m+y(f(x)-b); when the target value f(x) is greater than the threshold b, then the value of L is 0; the model is trained using the training set data, and the model parameters are adjusted to minimize the loss function so that the predicted value of the positive sample is greater than a certain threshold, while the predicted value of the negative sample is less than a certain threshold; then the convolution output values ​​dw[l] = dz[l]*a[l-1], db[l] = dz[l], da[l- 1]=W[l]T*dz[l], where da[l] is the input data, da[1] is the output data, and g[l]′ is the derivative of the activation function sigmoid; the back propagation of the convolution layer transfers the error term da[l]da[l] to the previous layer through a convolution operation, flips the convolution kernel and applies it to the error map, calculates the local gradient through the derivative of the activation function, and implements it through matrix operations, transforms the input and the convolution kernel into a matrix, and performs matrix multiplication, that is, pulls the numbers in the convolution window into a row to form a column vector, and performs matrix multiplication; the back propagation of the fully connected layer calculates the gradient of the weight and the gradient of the bias through the chain rule, and for each node, the error term is transferred to the node of the previous layer through the weight matrix, and calculates the local gradient through the derivative of the activation function, and implements it through matrix operations, multiplies the error term with the output of the current layer and the input of the previous layer, so that it is updated in the direction of minimizing the loss function.

[0156] S40, the model evaluation module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the model, and passes it to the parameter determination module;

[0157] See also Fig.12 As shown, the step S40 includes the following steps:

[0158] S41, the classification and marking unit divides the data sample into corresponding categories according to the quality abnormality categories of the LED driving power supply and marks them for the purpose of machine learning of the subsequent model, and transmits them to the learning and training unit;

[0159] To further illustrate, data classification is performed according to various quality anomalies of LED driver power supplies: "aging of electronic components" is category 1, "bad PCB" is category 2, "poor heat dissipation" is category 3, "unreasonable design" is category 4, "lightning damage" is category 5, "grid voltage fluctuation" is category 6, and "solder point failure" is category 7; when category 1 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (1,0,0,0,0,0,0); when category 2 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0,1,0,0,0,0,0); when category 3 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0,0,1,0,0,0,0); when category 4 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0,0 ,0,1,0,0,0), when category 5 is a positive sample, the other categories are negative samples, that is, the labeled data is (0,0,0,0,1,0,0), when category 6 is a positive sample, the other categories are negative samples, that is, the labeled data is (0,0,0,0,0,1,0), when category 7 is a positive sample, the other categories are negative samples, that is, the labeled data is (0,0,0,0,0,1,0); the above 6 abnormal categories are relatively typical abnormal classifications of LED driver power supplies, including but not limited to the above 7 categories, and one or more of the categories can also be re-divided into more categories, and a more detailed and in-depth division can be performed according to the actual production test results. The above classification is not fixed. The model is retrained after each category change to meet the actual needs of production.

[0160] S42, the learning and training unit determines the training model according to the gradient descent data, and inputs the data in the training set into the model for simulation training under different environmental conditions to ensure the accuracy of model recognition, and transmits it to the model evaluation unit;

[0161] It is further explained that the loss function is minimized and the prediction accuracy of the model is improved by continuously adjusting the model parameters to ensure that the model can accurately identify the adaptability of the LED driver power supply in different environments; the model is trained through the data set in the training set to fit the data analysis rules, that is, to determine the various learning parameters such as the model weight and bias; and the model parameters and hyperparameters are adjusted during the model training process through the data set in the validation set to optimize the model performance and avoid overfitting. The model is selected, but it does not participate in the determination of the learning parameters, and the model parameters and hyperparameters with smaller model errors are selected; then the generalization ability of the model on unknown data is evaluated through the data set in the test set after the model training is completed, and it is used once after training to evaluate the effect of the final model, and it does not participate in the learning parameter process or the hyperparameter selection process.

[0162] S43, the model evaluation unit obtains a comprehensive evaluation score F according to the model evaluation score calculation formula "F = 2 (Ac * Re) / (Ac + Re), F is the training model evaluation score, Ac is the precision, and Re is the recall rate", and transmits it to the processing center;

[0163] Further explanation, Ac and Re are respectively expressed by "Ac=(T P +T N ) / (T P +FP+FN+TN), Re=T P / (TP+FN)”, where T P is the number of true positive samples, F P is the actual number of false positive samples, F N is the number of false negative samples, T N is the number of true negative samples; after the model training is completed, a test is performed to verify the accuracy and reliability of the model. During the test, problems with the model may be found and optimized and adjusted; the training model evaluation score, namely the F value, is the harmonic mean of precision and recall, which can evaluate the performance of the classification model. The value range is 0 to 1, where 1 is the best performance and 0 is the worst performance. The higher the F value, the better the prediction effect of the model, and vice versa. Therefore, if the F value is close to 1, it indicates that the model performs well while maintaining a balance between precision and recall; adjust the hyperparameters according to the model's learning rate, regularization coefficient and other performance to optimize the model performance.

[0164] S44, the processing center compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is transmitted to the alarm and notified to continue training until it meets the standard.

[0165] It is further explained that the LED driver power quality abnormality category identification training model test evaluation score standard is best when it is greater than 0.8, which is specifically determined according to factors such as the protection function of the LED driver power supply, periodic aging test, conversion efficiency, power factor, electronic component aging, PCB quality, heat dissipation design, and power supply design; the power supply production process parameters include the production and testing equipment operating parameters of the LED driver power supply, power supply operation parameters, and production and testing procedures, among which the production and testing equipment operating parameters of the LED driver power supply include: the input voltage, power factor, protection function setting, layout and wiring rules of the automatic production line of the circuit board, and other parameters; the circuit board area and size of the automatic placement machine, the XY axis and Z axis movement range, the number of placement heads and the placement speed, the placement accuracy and the original size, the nozzle configuration and the feeding speed, the coordinate reference origin and the circuit board specifications, the program settings and calibration, and other parameters; the welding temperature, preheating zone temperature, welding time, wave crest height, wave crest speed, transmission speed, solder amount, and flux spray pressure of the automatic wave soldering machine. force, clamping angle and other parameters; welding temperature, preheating zone temperature, welding time, cooling time, temperature control accuracy, temperature uniformity, temperature curve repeatability, number of heating zones, heating speed, conveyor belt width, conveying speed, cooling efficiency and other parameters of automatic reflow soldering; input characteristics, output characteristics, stability characteristics, timing and transient characteristics, dimming frequency, dimming waveform, dimming cycle, input transient change characteristics and other test parameters of LED driver power tester; input characteristics, output characteristics, stability characteristics, timing and transient characteristics and other test parameters of LED driver power performance tester; temperature and humidity, test time, input / output characteristics, stability characteristics, timing and transient characteristics, protection characteristics and other test parameters of LED driver power aging test equipment; the operating parameters of the LED driver power supply include input voltage range, output voltage range, output current, power factor, working efficiency, working temperature, waterproof level, etc. These process parameters are automatically matched and applied in production through the trained model to avoid various quality abnormalities.

[0166] S50, the parameter determination module performs production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best production process parameters of the LED driver power supply to ensure the subsequent production quality, and passes them to the production application module;

[0167] See also Fig.13 As shown, the step S50 comprises the following steps:

[0168] S51. The power grid adjustment unit obtains the power grid adjustment rate according to the power grid adjustment rate calculation formula "Su=|Umax-Umin| / Uo*100%, Su is the power grid adjustment rate (%), Umax is the maximum output voltage (V) when the power input voltage changes, Umin is the minimum output voltage (V) when the power input voltage changes, and Uo is the rated output voltage (V) of the power supply" for subsequent power supply testing, and transmits it to the output ripple unit;

[0169] Further explanation, the grid regulation rate calculation formula "Su=|Umax-Umin| / Uo*100%" is used to measure the influence of input voltage change on output voltage when the load remains unchanged, and is one of the important indicators for evaluating the performance of LED driver power supply. The grid regulation rate is also called voltage regulation rate, which reflects the influence of input voltage change on output voltage when the load remains unchanged. The voltage regulation rate Su is proportional to the difference between the maximum output voltage Umax when the power input voltage changes and the minimum output voltage Umin when the power input voltage changes, and is inversely proportional to the rated output voltage Uo of the power supply; the maximum output voltage Umax when the power input voltage changes, the power input voltage Umin when the power input voltage changes, and the rated output voltage Uo when the power input voltage changes.

[0170] When the voltage changes, the minimum output voltage Umin and the rated output voltage Uo of the power supply are detected and obtained through the voltage sensor.

[0171] S52, the output ripple unit obtains the power supply output ripple according to the power supply output ripple calculation formula "Vr = Imax / (Co*f), Vr is the power supply output ripple (V), Imax is the maximum output current of the power supply (A), Co is the power supply output capacitance value (F), and f is the power supply switching frequency (Hz)" for subsequent power supply testing, and transmits it to the load adjustment unit;

[0172] Further explanation, the power supply output ripple calculation formula "Vr = Imax / (Co*f)" indicates that the output ripple voltage Vr is proportional to the maximum output current Imax, and inversely proportional to the output capacitance value (Co and the switching frequency f. Increasing the output capacitance value or increasing the switching frequency can reduce the ripple. In practical applications, the constant current source circuit is designed according to the rated working current of the LED lamp bead, so that the power supply output current is constant near the rated current of the lamp bead to ensure stable light emission of the LED and extend the service life; the power supply output ripple Vr is the fluctuation of the output voltage; the power supply maximum output current Imax, the power supply output capacitance value Co, and the power supply switching frequency f are obtained by referring to the power supply design parameters or through testing.

[0173] S53, the load adjustment unit calculates the power load adjustment rate according to the formula "L R =(U A -U B ) / U B *100%, LR is the power supply load regulation rate (%), U A UB is the output voltage (V) when the power supply is unloaded, and UB is the output voltage (V) when the power supply is fully loaded. The load regulation rate of the power supply is obtained for subsequent power supply testing and passed to the trial debugging unit;

[0174] Further explanation, the power supply load regulation rate calculation formula "LR = (UA-UB) / UB * 100%" reflects the ability of the power supply to maintain a stable output voltage when the load changes, and is an important indicator for evaluating the quality of the power supply. The power supply load regulation rate LR is proportional to the difference between the output voltage UA when the power supply is unloaded and the output voltage UB when the power supply is fully loaded, and is inversely proportional to the output voltage UB when the power supply is fully loaded; the output voltage UA when the power supply is unloaded is the output voltage of the LED driver power supply when no load is connected; the output voltage UB when the power supply is fully loaded is the output voltage when the LED driver power supply is connected to the maximum rated load; the output voltage UA when the power supply is unloaded and the output voltage UB when the power supply is fully loaded are obtained through detection by a voltage sensor.

[0175] S54, the trial debugging unit conducts trial production and testing of the LED driver power supply through the power supply production testing equipment according to the predetermined power supply production process parameters to obtain quality information to ensure the quality accuracy of the LED driver power supply, and transmits it to the processing center;

[0176] Further explanation: according to the control command, the prepared capacitors, inductors, diodes, MOSFET switch elements, rectifier bridges and other electronic components and necessary materials such as PCB boards, shells, heat sinks, etc. are subjected to laser cutting, electroplating, etching and other processes through the circuit board automatic production line and its predetermined operating parameters to produce each layer of the circuit board and ensure the accuracy and quality of the circuit board; then, through the automatic placement machine and its predetermined operating parameters, according to the design drawings of the LED drive power supply, the resistors, capacitors, diodes, ICs and other electronic components are accurately placed on the circuit board to improve production efficiency and accuracy; then, through automatic wave soldering, automatic reflow soldering and their predetermined operating parameters, the Wave soldering or reflow soldering and other processes are used to solder the mounted electronic components to the PCB board, and finally the heat sink, housing and other components are assembled with the PCB board through the power supply automatic assembly equipment to ensure the structural integrity of the power supply; then the assembled LED driver power supply is subjected to performance tests on indicators such as input voltage range, output voltage stability, load regulation rate, and grid regulation rate, and necessary debugging is performed based on the test results and compared with the calculation results of the above steps S51-S53 to ensure that the performance of the power supply meets the design requirements; the LED driver power supply that has undergone performance testing and debugging is quality checked to ensure that there is no damage or defects so as to facilitate subsequent normal production.

[0177] S55. The processing center compares the power quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is set as the formal production process parameter; if it does not meet the standard, it is transmitted to the alarm and notified to adjust the parameters for rework.

[0178] To further explain, the LED power supply production quality standards include: 1) In terms of safety standards: In North America, UL / ETL certification standards such as UL8750 and UL1310 must be followed, which stipulate the output voltage limit, circuit protection and other safety requirements of power supply equipment; in Europe, GS / CE / EMC certification standards such as EN61347-1 / -2-13, EN55015, EN61547, EN61000-3-2 / -3-3 must be met, which stipulate the general requirements and safety requirements of lamp control devices, as well as special requirements and electromagnetic compatibility standards for DC or AC electronic control devices for LED modules; In Australia, it is necessary to comply with SAA / C-Tick certification standards such as AS / NZS61347.1 / .2; in Japan, it is necessary to meet PSE safety regulations and EMC standards such as IEC / J55015 and IEC / J61347; in China, it is necessary to follow GB7000, GB17743, GB17625 and CCC / CQC standards such as GB19510.1 and GB19510.14 for LED modules, which stipulate the safety requirements and test methods for LED driver power supplies; 2) In terms of electromagnetic compatibility standards: mainly focus on GB / T17743 or CISPR for electrical lighting and similar equipment 15 and other radio disturbance characteristics limits and measurement methods, as well as electromagnetic compatibility limits such as GB17625.1 or IEC61000-3-2 harmonic current emission limits, to ensure that the LED driver power supply will not interfere with other electronic equipment during operation; 3) Follow certain inspection standards and sampling inspection plans to ensure that the power supply quality meets the specified requirements. When choosing an LED driver power supply, consumers should pay attention to whether it has passed relevant certifications, as well as factors such as brand reputation, to ensure that the quality of the purchased power supply is reliable.

[0179] S56. The abnormality recognition unit matches the real-time LED driver power production detection image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category for subsequent production improvement and parameter adjustment.

[0180] Further explanation: the image is input into the model and the trained model is used to predict the new monitoring image. The abnormal type of LED driver power supply is judged according to the output result. If the model's judgment on the abnormality of the LED driver power supply reaches a certain threshold, the early warning system is triggered and corresponding solutions are taken in time. The abnormal types of LED driver power supply quality include: 1) Aging of electronic components: due to the open circuit, short circuit, burnout, leakage, functional failure and other problems of resistors, capacitors, diodes, transistors, LEDs, connectors, ICs and other devices, high-quality electronic components should be selected to ensure that the components can work stably in a high temperature environment, and special attention should be paid to the selection of electrolytic capacitors and wires; 2) Poor PCB: due to the poor wetting, bursting, delamination, CAF (conductive ion migration), open circuit, short circuit and other problems of printed circuit boards (PCB) and printed circuit board assemblies (PCBA), the PCB design and manufacturing process should be optimized to ensure good wetting of the PCB and avoid problems such as delamination and CAF; 3) Poor heat dissipation: due to poor heat dissipation, the life of components such as electrolytic capacitors may be shortened, affecting the stability and life of LED lamps, and the heat dissipation design should be strengthened to ensure The power supply can effectively dissipate heat when working to avoid excessive temperature and shorten the life of components; 4) Power supply design problems: Due to insufficient power design, improper component selection, unreasonable electrical performance design (improper constant current parameter setting leads to inconsistent brightness, etc.), unreasonable PCB layout design, etc., the power supply parameters should be reasonably designed to ensure that the power design has sufficient margin. The component selection should consider the stability and life at high temperature. The electrical performance design should ensure the consistency of LED brightness and the stability of power supply output current; 5) Lightning damage: Since the transient waves generated by lightning strikes may have a fatal impact on the power supply line and cause damage to electronic components, lightning protection measures should be taken. Lightning protection measures should be considered during design to protect the power supply from damage by lightning surges; 6) Grid voltage fluctuations: Since the sharp fluctuations in grid voltage may exceed the load capacity of the power supply and cause damage to the driver, the grid voltage fluctuations should be monitored and corresponding protection measures should be taken to ensure that the power supply can work stably when the grid voltage fluctuates; 7) Solder joint failure: Due to solder joint quality problems, the reliability of the power supply may be reduced, which is caused by welding problems or thermal stress fatigue in the production process. The quality of the solder joints should be guaranteed to ensure the stability and durability of the solder joints and avoid power supply failures caused by welding problems.

[0181] S60, the production application module performs production and various tests on the LED driver power supply according to the set production process parameters and the LED driver power supply production and testing equipment according to the program instructions to ensure that the LED driver power supply meets the quality requirements.

[0182] See also Fig.14 As shown, the step S60 comprises the following steps:

[0183] S61, the substrate production unit produces each layer of the circuit board and mounts the electronic components on the circuit board through the automatic circuit board production line and the automatic placement machine according to the control command and the predetermined process parameters, and transmits the result to the mainboard assembly unit;

[0184] It is further explained that according to the control command, the prepared capacitors, inductors, diodes, MOSFET switching elements, rectifier bridges and other electronic components and necessary materials such as PCB boards, housings, heat sinks, etc. are subjected to laser cutting, electroplating, etching and other processes through the circuit board automatic production line and its predetermined input voltage, power factor, protection function settings, layout and wiring rules and other parameters to produce each layer of the circuit board and ensure the accuracy and quality of the circuit board; then, through the automatic placement machine and its predetermined circuit board area and size, XY axis and Z axis movement range, placement head quantity and placement speed, placement accuracy and original size, nozzle configuration and feeding speed, coordinate reference origin and circuit board specifications, program settings and calibration and other parameters, resistors, capacitors, diodes, ICs and other electronic components are accurately placed on the circuit board according to the design drawings of the LED drive power supply, thereby improving production efficiency and accuracy.

[0185] S62, the mainboard assembly unit solders the mounted electronic components to the PCB board through wave soldering or reflow soldering according to the control command and the predetermined process parameters, assembles the components with the heat sink and the housing into a power supply, and transmits the power supply to the appearance inspection unit;

[0186] It is further explained that according to the operating instructions, the mounted electronic components are welded together with the PCB board through automatic wave soldering and its predetermined welding temperature, preheating zone temperature, welding time, wave peak height, wave peak speed, transmission speed, solder amount, flux spray pressure, clamping inclination angle and other parameters or automatic reflow soldering and its predetermined welding temperature, preheating zone temperature, welding time, cooling time, temperature control accuracy, temperature uniformity, temperature curve repeatability, number of heating temperature zones, heating speed, conveyor belt width, conveying speed, cooling efficiency and other parameters through wave soldering or reflow soldering and other processes, and finally the heat sink, housing and other components are assembled with the PCB board through the power supply automatic assembly equipment, and the screw housing and the circuit board are locked together through the screw locking machine to ensure the stability and firmness of the circuit board, so as to make the structure of the LED driver power supply complete.

[0187] S63, the appearance inspection unit obtains the image or video of the power supply through the set high-definition camera and converts it into recognizable appearance quality information to ensure that the appearance quality of the power supply meets the customer requirements, and transmits it to the performance inspection unit;

[0188] Further explanation: according to the set process parameters, the high-definition camera is controlled to perform 360-degree shooting, and images of various positions of the LED power supply are taken from multiple different angles. The different positions of the LED power supply are identified at different angles, so that the positions and shapes of the LED power supply can be more accurately and comprehensively identified, or based on multiple images of the LED power supply taken from different angles, the LED power supply is identified by multi-angle matching, so as to determine whether the specifications and models of the printed board and components of the LED power supply meet the power supply requirements, whether the surface solder joints are smooth, whether the soldering is uniform, whether there are defects such as soldering, solder leakage, cold soldering and burrs, whether the appearance is clean, whether there is no oil stains and debris, whether the components on the printed board are neatly arranged, and whether the height and pin length meet the requirements, whether the shell has no obvious damage, deformation or burning marks, and whether the connection ends are smooth. Whether the port is loose, corroded or damaged, whether the color of the shell and input / output lines is uniform and there is no obvious color difference, whether the material is flame retardant, and whether the temperature resistance of each accessory is greater than 80 degrees, whether the logo is clear, to ensure the reliability and safety of the LED driver power supply during manufacturing and use; the power supply image includes: 1) real-time images during the production process, which can monitor the operating status of the production line and promptly discover and handle abnormal situations; 2) images of finished power supplies, to check the quality of the power supply for defects, color uniformity and other abnormalities; 3) images of raw materials, to ensure that the quality of raw materials meets production requirements and avoid production problems caused by raw material problems; according to specific production needs and algorithm requirements, it is also necessary to collect other types of image data such as images under different lighting conditions and images at different angles to improve the accuracy and robustness of the algorithm.

[0189] S64. The performance testing unit tests the circuit, mechanical, environmental reliability, electromagnetic compatibility, and transient protection performance of the power supply through the power supply performance testing equipment to ensure that the power supply quality meets the requirements and transmits it to the processing center;

[0190] To further explain, according to the control command, the assembled LED power supply is operated according to parameters such as input voltage range, output voltage range, output current, power factor, working efficiency, working temperature, waterproof level, etc., and the corresponding performance test is performed through the LED driver power supply tester and its predetermined input characteristics, output characteristics, stability characteristics, timing and transient characteristics, dimming frequency, dimming waveform, dimming cycle, input transient change characteristics and other test parameters; then, the relevant performance test is performed through the LED driver power supply performance tester and its predetermined input characteristics, output characteristics, stability characteristics, timing and transient characteristics and other test parameters; finally, the power supply aging test is performed through the LED driver power supply aging test equipment and its predetermined temperature and humidity, test time, input / output characteristics, stability characteristics, timing and transient characteristics, protection characteristics and other test parameters to ensure that the power supply meets the design requirements.

[0191] S65. The processing center compares the quality information of the power supply with the LED power supply production quality standard stored in the memory: if the standard is met, it is notified that the product can be shipped; if the standard is not met, it is transmitted to the alarm and notified to rework or repair.

[0192] Further explanation, in addition to the LED power supply production quality standards described in step S55 above, the LED power supply production quality standards also include: 1) Appearance: the selected printed board and component specifications must meet the power supply requirements; the solder joints are smooth and the solder is uniform, without defects such as solder joints, leaking solder joints, cold solder joints and burrs; the appearance is clean, free of oil stains and debris; the components on the printed board are arranged neatly, and the height and pin length must meet the regulations, that is, the height of the highest components such as the L1 inductor and Q1MOS tube on the printed board and the PCB should be less than 22mm, and the length of the pin should be less than 3mm; the shell has no obvious damage, deformation or burnt marks, and the connection port has no looseness, corrosion or damage to ensure a firm and reliable connection; the color of the shell and input / output lines should be uniform, without obvious color difference; the material must be flame retardant, and the temperature resistance of the shell, PVC wire and other accessories must be greater than 80 degrees; the surface has clear quality-approved signs / marks such as the Chinese CCC certification and the EU CE certification to ensure 1) The reliability and safety of LED driver power supply during manufacturing and use; 2) Performance: input surge current, output current, input power, working efficiency, output current ripple and impact current, power regulation rate, load regulation rate, temperature drift coefficient, temperature test, output overcurrent protection value and short circuit protection (OCP and SCP), output voltage overvoltage protection (OVP) and other circuit performance tests meet the relevant standards; metal shell, plastic (polymer) shell, input and output interface and brightness control interface, structure and other mechanical tests meet the standards of GB4208-2008; climate environment adaptability, vibration and other environmental reliability tests meet the standards of GB / T2423.1-2001, GB / T2423.2-2001, GB / T2423.3-1993; electromagnetic compatibility test meets the limits and measurement methods of radio disturbance characteristics of electrical lighting and similar equipment (GB / T17743 or CISPR 15), as well as electromagnetic compatibility limits and harmonic current emission limits (GB17625.1 or IEC61000-3-2), etc.; for the maximum value of the flicker index of the dimmable LED, as well as the transient protection characteristics and flicker index tests such as the ringing wave test, they meet the requirements of characteristic parameters such as transient protection characteristics and flicker index, thereby ensuring the quality and safety of the power supply.

[0193] It is further explained that the above-mentioned steps are displayed in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps. These steps can also be executed according to other orders, and some steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed and completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0194] Further explanation, the present invention is described according to the content implemented by the software program of the LED driver power supply production quality detection control system. Each LED driver power supply production quality detection method is divided into several modules or units to implement the software program instructions generated by each step. The software program instructions include the above-mentioned LED driver power supply production quality detection method.

[0195] An LED driver power supply production quality detection and control system provided by the present invention also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the LED driver power supply production quality detection method described in any one of the above are implemented; the computer-readable storage medium stores a computer program, and when the computer program is executed by the functional modules, the steps of the LED driver power supply production quality detection method described in any one of the above are implemented.

[0196] It is further explained that the computer-readable storage medium includes a memory, which can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program and its instructions corresponding to the LED driver power supply production quality detection method in the present invention, including information transmission instructions of each module; the memory executes various functional applications and data processing of each module by running the stored non-volatile software programs and instructions, that is, implements the LED driver power supply production quality detection method in the above-mentioned process embodiment; one or more units are stored in the memory, and when executed by the one or more modules, an LED driver power supply production quality detection method in any of the above-mentioned process embodiments is executed; the computer-readable storage medium stores computer executable instructions, which are executed by one or more modules, and can also be an LED driver power supply production quality detection method in any of the above-mentioned process embodiments.

[0197] To further illustrate, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two, and may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus or device, including but not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above, wherein the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, carrying a computer-readable program code, and the propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above, and may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable signal medium may send, propagate or transmit a program for use by or in combination with an instruction execution system, apparatus or device, and the program code that may be contained may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0198] It is further explained that the computer program is divided into multiple modules / units, which are stored in the memory and executed by each module / unit to complete the present invention; the multiple modules / units may be a series of computer program instructions that can complete specific functions, and the instructions are used to describe the execution process of the computer program in each module / unit.

[0199] It is further explained that the content described above in the present invention is written with reference to the implemented parts of the software copyrights such as "LED drive power supply circuit control system (registration number: 2018SR504415)" and "LED drive power supply dimming test system software (registration number: 2020SR1697325)" that the company has applied for successively since June 2018.

[0200] The present invention also provides an LED driving power supply detection control device, which is implemented by the above-mentioned LED driving power supply production quality detection method.

[0201] It is further explained that the control device is composed of the above-mentioned associated detection equipment or devices, and can be made into a complete set of LED power supply automatic production detection equipment together with the LED power supply production detection equipment according to the needs of the LED power supply manufacturer; it can also be made into a unit device of each functional module belonging to each step or the entire control device, and each control device is connected to each LED power supply production detection equipment through a wireless connection during installation; the structures of these devices are not described in detail here; the control device can also be used in professional LED power supply detection institutions to improve detection efficiency.

[0202] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application can be modified or replaced by equivalents, and all should be included in the scope of protection of the present application.

Claims

1. A method for detecting the production quality of an LED driver power supply, characterized in that: The method is applied to a production quality detection and control system for an LED driving power supply, the system comprising an information acquisition module, a model construction module, a model training module, a model evaluation module, a parameter determination module, a production application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal; the information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, the memory, and the alarm are respectively connected to the processing center; the intelligent mobile terminal is respectively connected to the wireless communication module wireless network within the range of a wireless network or the Internet; the method comprises the following steps: S10. Before production, the information acquisition module acquires normal and abnormal historical image data of LED driver power production and performs preprocessing for subsequent model construction, and passes it to the model construction module, including the following steps: S11. The data acquisition unit obtains historical data on circuit performance, mechanical properties, environmental reliability, electromagnetic compatibility, certification standards and production process parameters of similar LED power supplies by networking with the industry database, and transmits it to the cleaning data unit; S12, the cleaning data unit cleans the collected historical data to remove abnormal values, duplicate values ​​or missing values ​​in the data to improve the quality and accuracy of the data, and passes it to the enhanced data unit; S13, the enhanced data unit increases the quantity and diversity of LED power supply training data through data enhancement methods such as rotation, scaling, flipping, cropping, and color transformation to improve the generalization ability of the model, and passes it to the integrated data unit; S14, the integrated data unit obtains the standardized power data according to the data standardization processing formula "z = (x-μ) / σ, z is the standardized data, x is the original data, μ is the mean value of the data, σ is the standard deviation of the data" to improve the stability of the model, and passes it to the conversion data unit; S15, the conversion data unit obtains the normalized power data according to the normalization calculation formula "x'=(x-min(x)) / (max(x)-min(x)), x' is the normalized data, x is the original data, min(x) and max(x) are the minimum and maximum values ​​of the data respectively" to improve the model performance, and passes it to the filtering and denoising unit; S16, filtering and denoising unit is calculated according to the Gaussian filtering method" G(x,y) is the two-dimensional Gaussian function pixel, (x,y) is the pixel coordinate, and σ is the standard deviation” to obtain the denoised power image for grayscale image conversion and pass it to the grayscale conversion unit; S17, the grayscale conversion unit obtains the grayscale power image according to the grayscale calculation formula "f(i,j) = max(R(i,j), G(i,j), B(i,j)), f(i,j) is the grayscale image, R(i,j), G(i,j), B(i,j) are the original images of three colors respectively", so as to facilitate the subsequent feature extraction, and transmits it to the feature extraction unit; S18, the feature extraction unit converts the labeled data into a feature vector useful for model training and extracts parameter information, appearance structure characteristics, and key features of the protection mark on the label to improve the training speed and performance of the model; S20, the model building module determines the model architecture, designs the network hierarchy, adds auxiliary layers, and defines model parameters according to the extracted feature information to build the model for subsequent model training and verification, and passes it to the model training module; S30, the model training module trains the model according to the collected data and adjusts the model parameters to optimize the model performance to predetermine the power supply production process parameters, ensure the quality stability of the LED driving power supply, and pass it to the model evaluation module; S40, the model evaluation module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the model, and passes it to the parameter determination module; S50, the parameter determination module performs production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best power supply production process parameters to ensure the subsequent production quality, and transmits them to the production application module; S60, the production application module produces and performs various tests on the LED driver power supply according to the set production process parameters and the LED driver power supply production and testing equipment according to the program instructions to ensure that the LED driver power supply meets the quality requirements.

2. A method for detecting the production quality of an LED driving power supply according to claim 1, characterized in that : The system also includes: the wireless communication module is provided with a wireless network unit, which is responsible for sending and receiving wireless signals and automatically networking with the intelligent mobile terminal within the effective network range; The alarm compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory, and automatically sounds an alarm and notifies to continue training if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to adjust parameters for rework if the standard is not met; and compares the power quality information with the LED power supply production quality standard stored in the memory, and automatically sounds an alarm and notifies to rework or repair if the standard is not met; The memory is responsible for storing information of the information acquisition module, the model construction module, the model training module, the model evaluation module, the parameter determination module, the production application module, the wireless communication module, and the alarm, as well as the model evaluation sub-standard and the LED power supply production quality standard; The processing center is responsible for information transmission among functional modules, alarms, and memories, and is the hub of the system. It compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is passed to the alarm and notified to continue training; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is set as the formal production process parameter; if it does not meet the standard, it is passed to the alarm and notified to adjust the parameters for rework; and compares the power supply quality information with the LED power supply production quality standard stored in the memory: if it meets the standard, it is notified that it can be shipped; if it does not meet the standard, it is passed to the alarm and notified to rework or repair; The information acquisition module includes a data acquisition unit, a cleaning data unit, an enhancement data unit, an integration data unit, a filtering and denoising unit, a grayscale conversion unit, and a feature extraction unit, which is responsible for acquiring normal and abnormal historical image data of LED driver power production and preprocessing it, and passing it to the model building module; The model building module includes a network level unit, a network parameter unit, a hierarchical optimization unit, and a data partitioning unit, which are responsible for determining the model architecture, designing the network level structure, adding auxiliary layers, defining model parameters, and building the model according to the extracted feature information, and passing them to the model training module; The model training module includes an input filling unit, a convolution input unit, a pooling conversion unit, a fully connected layer unit, a normalized output unit, an output conversion unit, and a back-propagation unit, which is responsible for training the model according to the collected data and adjusting the model parameters to optimize the model performance to predetermine the power supply production process parameters, and pass them to the model evaluation module; The model evaluation module includes a classification labeling unit, a learning training unit, and a model evaluation unit, which is responsible for model verification and evaluation based on the trained model, and adjusting and optimizing the model based on the verification results, and passing it to the parameter determination module; The parameter determination module includes a power grid adjustment unit, an output ripple unit, a load adjustment unit, a trial debugging unit, and an abnormality identification unit, which is responsible for performing production tests under different conditions according to the production process parameters predicted by the trained model to obtain the best LED driver power supply production process parameters, and pass them to the production application module; The production application module includes a substrate production unit, a mainboard assembly unit, an appearance inspection unit, and a performance inspection unit, which are responsible for producing and inspecting the LED driver power supply according to the set production process parameters and program instructions of the LED driver power supply production inspection equipment to ensure that the LED driver power supply meets the quality requirements; The step S20 comprises the following steps: S21, the network level unit uses the selected convolutional neural network as the model architecture and customizes and optimizes the network level structure according to the specific application scenarios and requirements for subsequent defect detection and quality identification, and passes it to the network parameter unit; S22, the network parameter unit adjusts the number of network layers, convolution kernel size, step size, padding method and selects the loss function and optimization algorithm according to the network hierarchy structure to ensure the performance and accuracy of the model, and passes it to the hierarchical optimization unit; S23, the hierarchical optimization unit adjusts the size and normalization of the input data by adding a zero padding layer and a normalization layer before and after the convolution layer to improve the model training speed and stability, and passes it to the data partitioning unit; S24. The data partitioning unit divides the data set with key feature vectors into training set, validation set and test set according to the extraction and establishes a convolutional neural network structure for subsequent model training.

3. The LED driving power supply production quality detection method according to claim 1 is characterized in that: The step S30 comprises the following steps: S31, input filling unit according to data filling size calculation formula "PH = [(Ho-1)*SH + KH-Hi] / 2, PW = [(Wo-1)*SW + KW-Wi] / 2, PH and PW are the padding sizes in the height / width direction, Hi and Wi are the height / width of the input feature map, KH and KW are the height / width of the convolution kernel, SH and SW are the values ​​of the step length in the height / width direction, Ho and Wo are the height and width of the output feature map" to obtain the padding data and input it, and pass it to the convolution input unit; S32, the convolution input unit obtains the convolution input value and activates the function input according to the convolution input calculation formula "z(t) = ∫x(m)y(tm)dm, z(t) is the output function, x(t) is the input function, y(t) is the convolution kernel function, and dm is the integral differential element of the variable m", and passes it to the pooling conversion unit; S33, the pooling conversion unit obtains the maximum pixel value after pooling according to the maximum pooling calculation formula "Z(i,j)=max(X[i*PS(i+1)*PS,j*PS(j+1)*PS]), Z(i,j) is a pixel value in the output feature map after pooling, i and j are the positions in the output feature map, max is the maximum value of all pixel values ​​in the input window, X is the input feature map, PS is the window size of the pooling operation", enters the fully connected layer to retain important feature information, and passes it to the fully connected layer unit; S34, the fully connected layer unit obtains the output data of the connection layer according to the fully connected layer calculation formula "y=f(∑(wn*xn)+b), y is the output result, f is the activation function, wn is the weight of the nth input feature, xn is the nth input feature, n is the dimension of the input feature, and b is the bias" and enters the output layer for subsequent output calculation and is passed to the normalized output unit; S35, the normalized output unit obtains the output of Batch Norm according to the Batch Norm layer calculation formula "h = φ (BN (Wx + b)), h is the output of the fully connected layer, φ is the activation function, BN is the batch normalization operator, W is the weight parameter, x is the input of the fully connected layer, and b is the bias parameter", and passes it to the output conversion unit; S36, the output conversion unit obtains the output value size after convolution according to the output layer conversion calculation formula "N = (P-F + 2C) / S + 1, N is the output size after convolution, P is the input size before convolution, F is the convolution kernel size, C is the number of layers of 0 added around the image, and S is the step size" for subsequent model analysis and training, and passes it to the back propagation unit; S37, the back propagation unit obtains the loss function value according to the loss function calculation formula "L = max(0, m + y * (f (x) - b)), L is the loss function value, m is the hyperparameter, y is the sample label (0 or 1), f (x) is the predicted value of the model, and b is the classification boundary" and back propagates to optimize the performance of the network model; The step S40 comprises the following steps: S41, the classification and marking unit divides the data sample into corresponding categories according to the quality abnormality categories of the LED driving power supply and marks them for the purpose of machine learning of the subsequent model, and transmits them to the learning and training unit; S42, the learning training unit determines the training model according to the gradient descent data, and inputs the data in the training set into the model for simulation training under different environmental conditions to ensure the accuracy of model recognition, and transmits it to the model evaluation unit; S43, the model evaluation unit obtains the comprehensive evaluation score F according to the model evaluation score calculation formula "F = 2 (Ac * Re) / (Ac + Re), F is the training model evaluation score, Ac is the precision, Re is the recall rate", and transmits it to the processing center; S44, the processing center compares the actual evaluation score of the model with the evaluation score standard of the model stored in the memory: if it meets the standard, it is the predetermined power supply production process parameter; if it does not meet the standard, it is transmitted to the alarm and notified to continue training.

4. The LED driving power supply production quality detection method according to claim 1 is characterized in that: The step S50 comprises the following steps: S51. The power grid adjustment unit obtains the power grid adjustment rate according to the power grid adjustment rate calculation formula "Su=|Umax-Umin| / Uo*100%, Su is the power grid adjustment rate (%), Umax is the maximum output voltage (V) when the power input voltage changes, Umin is the minimum output voltage (V) when the power input voltage changes, and Uo is the rated output voltage (V) of the power supply" for subsequent power supply testing, and transmits it to the output ripple unit; S52, the output ripple unit obtains the power supply output ripple according to the power supply output ripple calculation formula "Vr = Imax / (Co*f), Vr is the power supply output ripple (V), Imax is the maximum output current of the power supply (A), Co is the power supply output capacitance value (F), and f is the power supply switching frequency (Hz)" for subsequent power supply testing and transmits it to the load adjustment unit; S53, the load adjustment unit obtains the power supply load adjustment rate according to the power supply load adjustment rate calculation formula "LR = (UA-UB) / UB * 100%, LR is the power supply load adjustment rate (%), UA is the output voltage of the power supply when there is no load (V), and UB is the output voltage of the power supply when there is full load (V)" for subsequent power supply testing, and transmits it to the trial debugging unit; S54, the trial debugging unit conducts trial production and testing of the LED power supply through the power supply production testing equipment according to the predetermined power supply production process parameters to obtain quality information to ensure the quality accuracy of the LED driving power supply, and transmits it to the processing center; S55. The processing center compares the power quality information with the LED power supply production quality standards stored in the memory: if the standards are met, they are set as formal production process parameters; if the standards are not met, they are transmitted to the alarm and notified to adjust the parameters for rework. S56, the abnormality recognition unit matches the real-time LED driver power production detection image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category for subsequent production improvement and parameter adjustment; The step S60 comprises the following steps: S61, the substrate production unit performs production of each layer of the circuit board and mounts electronic components on the circuit board through the automatic circuit board production line and the automatic placement machine according to the control command and the predetermined process parameters, and transmits the production to the mainboard assembly unit; S62, the mainboard assembly unit solders the mounted electronic components to the PCB board through wave soldering or reflow soldering according to the control command and the predetermined process parameters, and assembles the components into a power supply with the heat sink and the housing, and transmits the power supply to the appearance inspection unit; S63. The appearance inspection unit obtains the image or video of the power supply through the set high-definition camera and converts it into recognizable appearance quality information to ensure that the appearance quality of the power supply meets the customer requirements and transmits it to the performance inspection unit; S64. The performance testing unit uses power performance testing equipment to test the circuit, mechanical, environmental reliability, electromagnetic compatibility, and transient protection performance of the power supply to ensure that the power quality meets the requirements and transmits it to the processing center; S65. The processing center compares the quality information of the power supply with the LED power supply production quality standard stored in the memory: if it meets the standard, it will notify that it can be shipped; if it does not meet the standard, it will be transmitted to the alarm and notified to rework or repair.

5. A method for detecting production quality of an LED driving power supply according to claims 1-4, characterized in that: The system also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the LED driver power supply production quality detection method described in any one of claims 1 to 4 are implemented; the computer-readable storage medium stores a computer program, and when the computer program is executed by the functional modules, the steps of the LED driver power supply production quality detection method described in any one of claims 1 to 4 are implemented.

6. A LED drive power supply detection and control device, characterized in that: This is achieved by using a LED driver power supply production quality detection method as described in any one of claims 1 to 4 above.

Images