Intelligent integrated testing equipment for thermal fuse based on machine vision and neural network

Abstract

The application discloses a hot melt fuse intelligent integrated test equipment based on machine vision and neural network, which comprises a programmable power supply and an oven for installing a hot melt fuse connected through a circuit, and further comprises an intelligent integrated test instrument, a user interaction module and a machine vision module; the intelligent integrated test instrument is provided with a computer program; the computer program is used for adjusting parameters of the programmable power supply and the oven to establish a test circuit; the computer program is embedded with a learning control model based on neural network; the learning control model is based on instantaneous overload current, breaking current and experimental data of aging experiments of various hot melt fuses and overall experimental pictures obtained by the machine vision module to perform deep learning training, so that the test circuit required by experiments of various hot melt fuses is obtained and automatically updated. The application can realize automatic establishment of a hot melt fuse test circuit and improve test efficiency.

Description

Technical Field

[0001] This invention relates to an intelligent integrated testing device for thermal fuses based on machine vision and neural networks. Background Technology

[0002] A thermal fuse, also known as a thermal circuit breaker, is a non-resettable, one-time thermal protection device, mainly used for overheat protection in household appliances and industrial equipment.

[0003] Existing technologies already include mature intelligent drying oven products. These ovens combine thermal and refrigeration principles, incorporating both heating and cooling devices, and controlling the internal temperature through PID control and other methods. Simultaneously, mature high-precision programmable power supplies are also available. The functions and parameters of these power supplies can be controlled via computer software programming. For example, computer software programming can control the power supply's on / off state, adjust its output mode, output voltage, and output power value.

[0004] Testing thermal fuses is a crucial step in verifying their performance, and intelligent ovens can be used for this purpose. However, current thermal fuse testing mostly employs an on-site circuit setup method. This involves assembling the oven, power supply, switch, and wiring into a test circuit, adjusting the oven to the appropriate temperature, adjusting the power supply, and recording the test process and results. Currently, performance tests for a single product category, or for different product categories, are conducted separately and independently. After each experiment, researchers must manually record data and complete test logs. Before starting the next experiment, the test circuit must be rebuilt, consuming significant time and effort and increasing experimental costs. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent integrated testing device for thermal fuses based on machine vision and neural networks. By performing deep learning training on daily data and experimental images of various tests of thermal fuses, a learning control model is obtained, enabling the automatic construction of thermal fuse test circuits and realizing intelligent automated testing of thermal fuses.

[0006] This invention discloses an intelligent integrated testing device for thermal fuses based on machine vision and neural networks. The device includes a programmable power supply and an oven for mounting thermal fuses, all connected by a circuit. It also includes an intelligent integrated testing instrument, a user interaction module, and a machine vision module. The intelligent integrated testing instrument is equipped with a computer program. The computer program adjusts the parameters of the programmable power supply and the oven to establish a test circuit. Furthermore, the computer program embeds a neural network-based learning control model. This model undergoes deep learning training based on experimental data from various thermal fuse experiments, including instantaneous overload current, breaking current, and aging tests, as well as the overall experimental footage acquired by the machine vision module. This training results in the acquisition and automatic updating of the test circuits required for various thermal fuse experiments.

[0007] This invention employs a programmable power supply to acquire its own status in real time, including the current actual voltage, current actual current, power supply operating mode, and power switch status, and uploads this information to an intelligent integrated testing instrument. The learning control model, based on the acquired test requirements, specifically, allows the user interaction module to scan the test parameter table via the machine vision module or manually input test parameters to automatically build the test circuit. The process involves the intelligent integrated testing instrument sending control commands to the programmable power supply to modify various parameters such as voltage, current, and operating mode. Simultaneously, the intelligent integrated testing instrument reads and records the oven temperature and sends control commands corresponding to the test requirements to the oven. After installing the thermal fuse inside the oven, testing can be performed using the automatically built test circuit. This invention integrates test circuits for instantaneous overload current, breaking current, and aging tests of thermal fuses.

[0008] The present invention also has the following preferred designs:

[0009] The learning control model of the computer program of the present invention selects a feedforward neural network and a convolutional neural network for learning and training. The feedforward neural network uses a BP neural network with an input layer + 8 hidden layers + an output layer structure. The input data of the input layer are voltage, current, power, power factor and temperature. The classic tanh hyperbolic tangent transfer function is used. The convolutional neural network processes the overall experimental image obtained by the machine vision module, adopts the traditional YOLOv3 model, and uses multi-scale features for object detection.

[0010] Preferably, the machine vision module uses a dual-camera setup consisting of a high-definition camera and an infrared camera. The training of the learning control model employs two convolutional neural networks, and a backpropagation (BP) neural network with one input layer, two hidden layers, and one output layer can be added to the output layer of the two BP neural networks to comprehensively judge the results and output a current result. The learning control model uses previous experimental data and experimental images as data samples for neural network training.

[0011] The test circuit of the present invention consists of at least a DC pulse source, a programmable DC power supply, a programmable AC power supply, an AC / DC electronic load, an adjustable load, a relay, and a switch. The relay communicates with the PLC controller to control the on / off state of each switch and to construct the test circuit between the power supply, the load, and the thermal fuse.

[0012] Preferably, the adjustable load includes an adjustable resistor, an adjustable motor load, and an adjustable inductor.

[0013] Based on the experimental testing requirements, this invention, when the thermal fuse is an AC type, disconnects the DC pulse source and the programmable DC power supply, and uses a programmable AC power supply; similarly, when the thermal fuse is a DC type, a programmable DC power supply is used; and similarly, when the thermal fuse is a DC pulse type, a DC pulse source is used. The required load type is tested by controlling the on / off state of the thermal fuse using a relay-controlled switch.

[0014] The intelligent integrated testing instrument of the present invention acquires voltage, current, power, power factor, and temperature, and determines whether the voltage, current, power, power factor, and temperature exceed the specified range during the experiment. If they exceed the specified range, the experiment is determined to be abnormal, and the instrument also determines whether the experimental image acquired by the machine vision module is abnormal.

[0015] When each experimental project of the present invention is completed, the experimental data corresponding to the experimental project is acquired, and the overall experimental screen of the experimental process is acquired through the machine vision module. The voltage, current, power, power factor, temperature and experimental screen collected during the experimental project are stored as new sample data. The stored sample data of multiple experimental projects are input into the neural network for retraining and updating the version of the learning training model. Higher weights can be assigned to the new sample data. Through weight training, the accuracy of the updated version depends more on the judgment result of the updated data.

[0016] The sample data input to the neural network in this invention includes experimental anomaly data and abnormal experimental images for each experimental project. This allows the learning control model to comprehensively judge the current experimental status and facilitate timely handling of experimental anomalies. For example, when the pre-trained neural network detects an anomaly, it immediately sends a control command to attempt to eliminate the anomaly. If the anomaly elimination fails, the experiment is terminated, and the testing engineer is notified of the end of the experiment via the interactive interface. After each experiment, the intelligent integrated testing instrument can summarize all experimental data and generate a data summary document and a report document based on the report template and data summary template, which are then sent to the testing engineer.

[0017] The infrared camera of the machine vision module of the present invention is equipped with a temperature limit. When the local temperature of the monitored oven, intelligent integrated testing instrument, etc. exceeds the temperature limit, relevant safety measures or warnings can be activated.

[0018] The beneficial effects of this invention are as follows:

[0019] 1. This invention provides a learning control model for testing thermal fuses based on machine vision and neural networks. It can automatically build test circuits for various thermal fuse instantaneous overload current, breaking current, and aging tests according to the experimental testing requirements, avoiding various losses from rebuilding test circuits, improving test results, and significantly reducing test costs.

[0020] 2. This invention uses experimental data and the overall experimental process as sample data, and uses a neural network to train a learning control model, which can comprehensively judge the current experimental status and facilitate timely handling of experimental anomalies.

[0021] 3. The present invention uses the data from each experiment and the experimental images as new sample data for neural network training, thereby enabling the learning control model to self-update and increasing the accuracy of the system over time. Attached Figure Description

[0022] Figure 1 This is a block diagram illustrating the working principle of an intelligent integrated testing device for thermal fuses based on machine vision and neural networks according to the present invention.

[0023] Figure 2 This is a schematic diagram of the neural network principle of the learning control model of the present invention;

[0024] Figure 3 This is a network structure diagram of the open-source convolutional neural network YOLOv3;

[0025] Figure 4 This is a schematic diagram illustrating the principle of adding a backpropagation (BP) neural network to the output layer of the convolutional neural network in this embodiment.

[0026] Figure 5 This is a schematic diagram of the electrical circuit of an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the adjustable inductor according to an embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments, so that those skilled in the art can better understand and implement the technical solution of the present invention.

[0029] like Figures 1 to 6As shown, an intelligent integrated testing device for thermal fuses based on machine vision and neural networks includes a programmable power supply and an oven for mounting thermal fuses connected by a circuit. It also includes an intelligent integrated testing instrument, a user interaction module, and a machine vision module. The intelligent integrated testing instrument is equipped with a computer program. The computer program adjusts the parameters of the programmable power supply and the oven to establish a test circuit. Furthermore, the computer program embeds a neural network-based learning control model. This learning control model is trained through deep learning based on experimental data from various thermal fuses, including instantaneous overload current, breaking current, and aging tests, as well as the overall experimental images acquired by the machine vision module. This allows it to obtain and automatically update the test circuits required for various thermal fuse experiments.

[0030] The programmable power supply acquires its own status in real time, including the current actual voltage, current actual current, power supply operating mode, and power switch status, and uploads this information to the intelligent integrated testing instrument. Based on the acquired test requirements, the learning control model, specifically, uses the machine vision module to scan the test parameter table or manually inputs test parameters to automatically build the test circuit. During the test circuit construction process, the intelligent integrated testing instrument sends control commands to the programmable power supply, modifying various parameters such as voltage, current, and operating mode. Simultaneously, the intelligent integrated testing instrument reads and records the oven temperature and sends control commands corresponding to the test requirements to the oven. After installing the thermal fuse inside the oven, testing can be performed using the automatically built test circuit. This integrates test circuits for thermal fuse instantaneous overload current, breaking current, and aging tests.

[0031] As a preferred embodiment:

[0032] like Figure 2 As shown, the learning control model of the computer program selects a feedforward neural network and a convolutional neural network for learning and training. The feedforward neural network uses a BP neural network with an input layer + 8 hidden layers + an output layer structure. It uses previous experimental data and experimental images as data samples for neural network training. The input data of the input layer are voltage, current, power, power factor and temperature, and the classic tanh hyperbolic tangent transfer function is used. The convolutional neural network processes the overall experimental images acquired by the machine vision module, adopts the traditional YOLOv3 model, and uses multi-scale features for object detection.

[0033] Preferably, the machine vision module employs a dual-camera setup consisting of a high-definition camera and an infrared camera. The training of the learning and control model utilizes two convolutional neural networks (CNNs), and a backpropagation (BP) neural network with one input layer, two hidden layers, and one output layer can be added to the output layer of these two CNNs. This BP neural network is used for image data processing of the overall experimental footage acquired by the cameras and is responsible for ensuring the safety of the experimental environment. The CNNs are modified from the open-source code YOLOv3, which is licensed under the MIT license. Its network structure is as follows: Figure 3 As shown, convolutional layers are convolutional layers, and residual layers are residual layers. They use convolutional kernels of corresponding sizes for convolution and feature boxes for prediction. The final output is a feature box containing the location, size, and prediction accuracy probability.

[0034] Based on this, most results with low accuracy are discarded, and only three feature boxes with an accuracy of over 90% are selected. If none are found, they are filled in with empty data. The positions of the output feature boxes and their predicted accuracy probabilities are used as input data to a backpropagation (BP) neural network for classification. This BP neural network uses a 1-layer input layer, 2-layer hidden layers, and 1-layer output layer, and is trained using a large amount of manual judgment results and the aforementioned data. Figure 4 As shown, the final output is mapped to five categories of results: experimental abnormality, serious experimental abnormality, normal experimental progress, experimental completion, and unexpected unknown situation in the experiment. The results can be judged comprehensively.

[0035] The intelligent integrated testing instrument acquires voltage, current, power, power factor, and temperature. It determines whether the voltage, current, power, power factor, and temperature exceed the specified range during the experiment. If they exceed the specified range, the experiment is deemed abnormal. The instrument also determines whether the experimental image acquired by the machine vision module is abnormal. The output layer outputs the current experimental status, with 0 representing abnormality and 1 representing normality.

[0036] Upon completion of each experimental project, the corresponding experimental data is acquired, and the entire experimental device is covered by the field of view of the machine vision module to obtain the overall experimental picture of the experimental process. The voltage, current, power, power factor, temperature, and experimental picture collected during the experimental project are stored as new sample data. The stored sample data of multiple experimental projects are input into the neural network for retraining and updating the version of the learning training model. Higher weights can be assigned to the new sample data. Through weight training, the accuracy of the updated version depends more on the judgment results of the updated data.

[0037] The neural network's sample data includes experimental anomaly data and abnormal experimental images for each experiment. This allows the learning control model to comprehensively judge the current experimental status and facilitate timely handling of experimental anomalies. For example, when the pre-trained neural network detects an anomaly, it immediately sends a control command to attempt to eliminate the anomaly. If the anomaly elimination fails, the experiment is terminated, and the testing engineer is notified of the end of the experiment via the interactive interface. After each experiment, the intelligent integrated testing instrument can summarize all experimental data and generate a data summary document and a report document based on the report template and data summary template, which are then sent to the testing engineer.

[0038] The infrared camera of the machine vision module is equipped with a temperature limit. When the local temperature of the monitored oven, intelligent integrated testing instrument, etc. exceeds the temperature limit, relevant safety measures or warnings can be activated.

[0039] like Figure 5 and 6 As shown

[0040] In this embodiment, the test circuit consists of at least a DC pulse source, a programmable DC power supply, a programmable AC power supply, an AC / DC electronic load, an adjustable load, relays, and switches. The relays communicate with the PLC controller to control the on / off switching of each switch. The test circuit between the power supply, load, and thermal fuse is constructed to meet the requirements of thermal fuse testing. The test conditions include: corresponding load type, current value, voltage value, power factor, and current characteristics.

[0041] Specifically, the adjustable load includes an adjustable resistor R1, an adjustable motor M1, and an adjustable inductor L1. The computer program of the intelligent integrated testing instrument determines the load and communicates with the PLC controller. By controlling the output of the PLC controller corresponding to the testing requirements, it controls the corresponding relays to control the on / off state of each switch, thus realizing the construction of the testing circuit. See [link to relevant documentation]. Figure 5 Different adjustable loads can be selected by switching on and off the bidirectional switches K2, K3, and K4, and different power supplies can be selected by switching K6, K7, and K8.

[0042] When testing a thermal fuse and there is a power factor requirement, the adjustable inductor L1 is connected via switch K4, such as... Figure 6As shown, the adjustable inductor L1 relies on 11 sets of relays to control the on / off state of corresponding switches K12 to K22, adjusting the inductance value. The adjustable inductance values ​​are 8H, 16H, 32H, 64H, 128H, 256H, 512H, 1024H, 2048H, 4096H, and 8192H. Based on years of testing experience, there is virtually no need to test inductors below 8H, nor is there a need for inductors above 16384H. Therefore, setting 11 inductance values ​​is sufficient to cover most application scenarios. The adjustable resistive load R1 can be set to 0.2Ω, 0.5Ω, 1Ω, 2Ω, 4Ω, and 8Ω. Under normal circumstances, the adjustable resistor R1 is kept open, and AC / DC electronic loads can meet the requirements for thermal fuse testing. The adjustable resistor R1 is used for fine-tuning.

[0043] According to the test requirements of the experiment, when the thermal fuse is an AC type thermal fuse, the DC pulse source and the programmable DC power supply should be disconnected, and the programmable AC power supply should be used; when the thermal fuse is a DC type thermal fuse, the programmable DC power supply should be used in the same way; when the thermal fuse is a DC pulse type thermal fuse, the DC pulse source should be used in the same way.

[0044] The above embodiments are merely preferred embodiments of the present invention, but should not be construed as limiting the invention. Any modifications and improvements made based on the concept of the present invention should fall within the protection scope of the present invention, and the specific protection scope is subject to the claims.

Claims

1. A smart integrated testing device for thermal fuses based on machine vision and neural networks, comprising a programmable power supply and an oven for mounting thermal fuses connected by circuitry, characterized in that, It also includes an intelligent integrated testing instrument, a user interaction module, and a machine vision module. The intelligent integrated testing instrument is equipped with a computer program. The computer program adjusts the parameters of the programmable power supply and the oven to establish a test circuit. The computer program embeds a neural network-based learning control model. The learning control model is trained by deep learning based on experimental data of instantaneous overload current, breaking current, and aging tests of various thermal fuses, as well as the overall experimental images obtained by the machine vision module, to obtain and automatically update the test circuit required for various thermal fuse experiments. The test circuit consists of at least a DC pulse source, a programmable DC power supply, a programmable AC power supply, an AC / DC electronic load, an adjustable load, a relay, and a switch. The relay communicates with the PLC controller to control the on / off state of each switch and to construct the test circuit between the power supply, the load, and the thermal fuse. The learning control model of the computer program selects a feedforward neural network and a convolutional neural network for learning and training. The feedforward neural network uses a BP neural network with an input layer, 8 hidden layers, and an output layer. The input data of the input layer are voltage, current, power, power factor, and temperature. The classic tanh hyperbolic tangent transfer function is used. The convolutional neural network processes the overall experimental image acquired by the machine vision module, using the traditional YOLOv3 model and multi-scale features for object detection.

2. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 1, characterized in that: The machine vision module uses a dual-camera setup consisting of a high-definition camera and an infrared camera, and the learning control model is trained using two convolutional neural networks.

3. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 2, characterized in that: The adjustable load includes an adjustable resistor, an adjustable motor load, and an adjustable inductor.

4. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 3, characterized in that: The intelligent integrated testing instrument acquires voltage, current, power, power factor, and temperature, and determines whether the voltage, current, power, power factor, and temperature exceed the specified range during the experiment. If they exceed the specified range, the experiment is determined to be abnormal, and the instrument also determines whether the experimental image acquired by the machine vision module is abnormal.

5. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 4, characterized in that: Upon completion of each experimental project, the corresponding experimental data is acquired, and the overall experimental process is captured through the machine vision module. The voltage, current, power, power factor, temperature, and experimental images collected during the experimental project are stored as new sample data. The stored sample data from multiple experimental projects are input into the neural network for retraining, updating the version of the learning and training model.

6. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 5, characterized in that: The sample data input to the neural network includes experimental anomaly data and abnormal experimental images for each experimental project.

7. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 6, characterized in that: The infrared camera in the machine vision module is equipped with a temperature limit.

8. The intelligent integrated testing equipment for thermal fuses based on machine vision and neural networks according to claim 1, characterized in that: The user interaction module obtains test requirements and automatically builds the test circuit by scanning the test parameter table through the machine vision module or by manually entering test parameters.

Images