Ultrasonic wire detection method for display-like devices

Abstract

The application provides an ultrasonic line detection method for display devices, comprising the following steps: S1, for a normal display panel, an ultrasonic generator is used to emit ultrasonic waves and record echo signals of the normal panel, and a normal echo signal dataset is established; S2, for a defect display panel detection analysis sample, an ultrasonic generator is used to emit ultrasonic waves and record echo signals corresponding to defect positions, and a defect signal dataset is formed; S3, the echo signals of the defect positions are compared with normal echo signals at the same positions, and the defect type of the defect analysis sample is marked, and a line defect dataset is obtained according to the echo signals corresponding to the defect positions obtained by the defect display panel detection analysis sample, the line defect dataset is preprocessed and noise data is added, and a display device line defect judgment model is obtained by training the line defect training set through a deep learning model; the detection precision is high and the detection efficiency is high.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology for display devices, and in particular to an ultrasonic circuit testing method for display devices. Background Technology

[0002] Currently, the quality of the display screen plays a crucial role in the overall quality of display devices. However, the detection of defects in screen traces in display panels, especially in LCD panels, is usually achieved through circuit testing and AOI equipment. In LCD panels, the detection of defective products often requires manual disassembly of the upper and lower substrates for fragmentation before further inspection of the circuitry using a camera. This process is time-consuming and labor-intensive, and the fragmentation process is irreversible, making it impossible to inspect the overall performance of the entire module after fragmentation.

[0003] To address the aforementioned issues, Chinese Patent Application No. 202210259462.X, published on July 29, 2022, discloses an AOI-based system and method for detecting defects in bare PCB boards. Specifically, it discloses a system comprising a camera positioning module, an image acquisition module, an image positioning module, an image preprocessing module, an image feature extraction module, a target recognition module, an image transmission module, and a result output and display module. The camera positioning module is used for positioning the camera in three-dimensional coordinates; the image positioning module is used for precise positioning of the PCB board to be inspected. This document uses two cameras simultaneously to locate, acquire, and preprocess images of the bare PCB board to be inspected, allowing defects to be separated from the image. This enables the detection of defects in PCB board trace width and spacing, as well as short circuits, open circuits, voids, dents, protrusions, and scratches. Furthermore, the defects on the PCB board are categorized and statistically analyzed according to circuit layers, carbon wire layers, solder pad layers, and mechanical layers, facilitating user review and monitoring. This allows for rapid problem detection and demonstrates high practicality.

[0004] This method requires two cameras to locate the defect. However, for circuit defects inside the display panel, the circuit board needs to be removed from the display panel for testing. This testing method is unreliable. Furthermore, the detection of micro-circuits by camera requires high camera accuracy, and the camera detection accuracy is not high. Summary of the Invention

[0005] The purpose of this invention is to provide an ultrasonic circuit testing method for display devices, which can classify and label different defects, resulting in high testing efficiency and more accurate circuit testing.

[0006] To achieve the above objectives, the present invention provides an ultrasonic circuit testing method for display devices, comprising the following steps: S1 uses an ultrasonic generator to emit ultrasonic waves and record the echo signals of the normal display panel to establish a normal echo signal dataset.

[0007] S2 analyzes defect display panel samples by using an ultrasonic generator to emit ultrasonic waves and record the echo signals corresponding to the defect locations, thus forming a defect signal dataset.

[0008] S3 compares the echo signal at the defect location with the normal echo signal at the same location, marks the defect type for the defect analysis sample, and obtains the line defect dataset based on the echo signal corresponding to the defect location obtained from the defect display panel detection analysis sample. The line defect dataset is preprocessed and noise data is added. A deep learning model is used to train the line defect training set to obtain the line defect judgment model for display devices.

[0009] S4 performs ultrasonic circuit testing on the abnormal display panel, and uses a circuit defect judgment model for display devices to detect circuit abnormalities in the area to be tested on the abnormal screen. It also calculates whether the ultrasonic echo signal parameters of the abnormal area meet the preset detection accuracy. If so,... Proceed to step S41: Input the ultrasonic echo signal of the abnormal area into the display device circuit judgment model to judge the defect; if not, proceed to step S42: Add a preset amount of noise data to the circuit defect training set and then re-enter step S4 to perform detection until the preset detection accuracy is reached.

[0010] S5 uses the display device circuit defect judgment model to determine whether the first detection point in the area to be detected is a defect. If it is, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to judge the area defect. If not, the detection of the first detection point in the area to be detected ends.

[0011] S6 confirms whether all areas to be inspected have been measured. If yes, end; otherwise, move the mobile platform to the next to be inspected point to judge defects until all inspection points have been inspected.

[0012] The above method establishes a dataset of normal ultrasonic echo signals and defective ultrasonic waves at corresponding coordinate positions of display panels of the same model. It compares the echo signals at defective locations with the normal echo signals to identify circuit abnormalities from abnormal echo detection signals. For abnormal echo signals, corresponding labels are assigned, and an abnormal echo signal dataset is established. Noise is added to the abnormal echo signal dataset to enhance the abnormal echo signals, improving the generalization ability of the dataset and the model. Finally, a deep learning model is used to train the abnormal echo signal dataset, resulting in a circuit defect judgment model for display devices. This model is then used to perform ultrasonic circuit detection on abnormal display panels to confirm whether the abnormal echo signals meet the defined detection accuracy. If the detection accuracy is met, the abnormal ultrasonic echo signals are input into the display device circuit. The defect judgment model, without artificially adding noise, continues to train the model to make the defect ultrasonic dataset more closely resemble the normal ultrasonic dataset, thereby achieving the preset detection accuracy and improving detection precision. Then, the display device circuit defect judgment model obtained from the training set is used to determine whether the first detection point in the area to be inspected is defective. If so, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to determine the defect in the area. If not, the detection of the first detection point ends. It is then confirmed whether all areas to be inspected have been measured. If so, the process ends; otherwise, the moving platform moves to the next to be inspected point to judge defects until all detection points have been inspected. Thus, all inspections can be performed using only the obtained display device circuit defect judgment model, resulting in high detection efficiency.

[0013] Furthermore, in step S1, the normal display panel is moved to a designated position above the normal display panel by a moving platform, the movement of which is controlled by a control motor.

[0014] With the above settings, the motor rotation controls the spatial coordinates of the end effector to detect the position to be detected, enabling accurate and rapid arrival at the detection position and improving detection efficiency.

[0015] Furthermore, in step S2, the defect display panel detection and analysis sample is moved to above the defect location by a moving platform, the movement of which is controlled by a control motor.

[0016] The above setup is made mobile via a mobile platform, and its structure is simple and reliable.

[0017] Furthermore, in step S3, the defect type is marked, and the ultrasonic signal segment of the defect area is marked. The markings are glass crack, micro crack, and broken wire.

[0018] The above settings categorize and label defect types, making it easier to organize defect types and subsequently determine the defect type directly from the corresponding echo signal.

[0019] Furthermore, the defect type labeling for the defect analysis samples includes: analyzing the first parameter of the collected echo signal, comparing the differences in the first parameter, and using software to label the different first parameters in the defect sample.

[0020] The above settings allow for comparison and annotation of echo signals, making the data more intuitive and enabling more direct identification of defect types.

[0021] Furthermore, the first parameter includes the amplitude and frequency of the ultrasound.

[0022] The above settings, since amplitude and frequency best reflect the characteristics of ultrasonic signals, enable more accurate marking.

[0023] Furthermore, the echo signal set is annotated with the first parameter of the ultrasonic signal using Praat software to form a normal echo signal dataset and a defect signal dataset.

[0024] The above settings categorize the echo signals, facilitating subsequent comparison and organization.

[0025] Furthermore, noise data is added to a portion of the randomly selected line defect dataset to introduce artificial noise. This artificial noise is the noise generated during production line operation, or the typical noise inherent in the equipment itself, or randomly selected operational noise.

[0026] The above settings enhance data diversity and expand dataset size by adding noise, thereby increasing the model's generalization ability and robustness. Preprocessing also makes training more accurate and efficient.

[0027] Furthermore, the deep learning model is the YOLOv5s model.

[0028] The above settings divide the YOLOv5s model dataset into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used for adjusting model parameters and thresholds, and the test set is used for evaluating model performance and generalization ability. The dataset is labeled and divided to facilitate model training, parameter adjustment, and performance evaluation.

[0029] Furthermore, the YOLOv5s model requires modification of the dataset configuration file and the YOLOv5s model configuration file. The pre-trained weight file is placed in the specified path of the weights, and the parameters of the weight path are updated accordingly. The parameters and file paths of the modified model YOLOv5_panel.yaml and the dataset panel.yaml are updated. Finally, the YOLOv5s model is trained in the system.

[0030] The above settings update the data by updating the configuration file, ensuring that the data is updated before the predetermined accuracy is reached, thus guaranteeing that the detection results meet the requirements. Attached Figure Description

[0031] Figure 1 This is a flowchart of the present invention.

[0032] Figure 2 This is a flowchart of the detection steps of the present invention.

[0033] Figure 3 The flowchart illustrates the detection process for this invention. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0035] like Figures 1 to 3 As shown, the present invention provides an ultrasonic circuit detection method for display devices. The testing method includes the following steps: S1 uses an ultrasonic generator to emit ultrasonic waves and record the echo signals of the normal display panel to establish a normal echo signal dataset.

[0036] S2 analyzes defect display panel samples by using an ultrasonic generator to emit ultrasonic waves and record the echo signals corresponding to the defect locations, thus forming a defect signal dataset.

[0037] S3 compares the echo signal at the defect location with the normal echo signal at the same location, marks the defect type for the defect analysis sample, and obtains the line defect dataset based on the echo signal corresponding to the defect location obtained from the defect display panel detection analysis sample. The line defect dataset is preprocessed and noise data is added. A deep learning model is used to train the line defect training set to obtain the line defect judgment model for display devices.

[0038] S4 performs ultrasonic circuit testing on the abnormal display panel, and uses a circuit defect judgment model for display devices to detect circuit abnormalities in the area to be tested on the abnormal screen. It also calculates whether the ultrasonic echo signal parameters of the abnormal area meet the preset detection accuracy. If so,... Proceed to step S41: Input the ultrasonic echo signal of the abnormal area into the display device circuit judgment model to judge the defect; if not, proceed to step S42: Add a preset amount of noise data to the circuit defect training set and then re-enter step S4 to perform detection until the preset detection accuracy is reached.

[0039] S5 uses the display device circuit defect judgment model to determine whether the first detection point in the area to be detected is a defect. If it is, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to judge the area defect and judge the accuracy. If not, the detection of the first detection point in the area to be detected ends.

[0040] S6 confirms whether all areas to be inspected have been measured. If yes, end; otherwise, move the mobile platform to the next to be inspected point to judge defects until all inspection points have been inspected.

[0041] In the above steps, normal data and defect data are collected through a mobile platform. Then, the defect data is classified and labeled, and enhanced preprocessing is performed. Then, a defect judgment model for display device circuits is obtained through a deep learning model. Finally, the area to be detected is judged based on the defect judgment model for display device circuits.

[0042] In one embodiment, in S1, normal samples are detected using photoultrasound technology based on the spatial coordinates of the control end of the motor rotation control. The received echo signals are converted into electrical signals using an ultrasonic transducer, and finally, a normal echo signal dataset is established by labeling with Praat software.

[0043] In S2, the spatial coordinates of the control end of the motor rotation are used to detect abnormal samples using photoultrasound. The received echo signals are converted into electrical signals using an ultrasonic transducer. Finally, the normal echo signal dataset is established by labeling with Praat software.

[0044] In S3, the first parameter of the collected ultrasonic signals is analyzed and compared. The differences in the first parameter, such as amplitude and frequency, are compared. The differences in parameters in the defect samples are marked by software. Then, the ultrasonic signal at the defect location is compared with the normal echo signal at the same location. The defect type is labeled for the defect analysis samples, and a dataset is established. Data preprocessing involves two steps: First, data cleaning, which involves screening ultrasonic images and removing those that do not meet the requirements. Specifically, the images are compared with preset correct images to determine whether they meet the requirements, ensuring the quality and reliability of the dataset. Second, data augmentation is performed, which involves artificially adding noise to some samples in the dataset. The noise can be the noise from the production line, the typical noise of the equipment itself, or random cropping, etc. Noise is introduced by adding a preset value, such as 1 dB. Then, a deep learning model is trained. This step involves dataset labeling and partitioning, as well as model training. To facilitate model training, parameter adjustment, and performance evaluation, the dataset is partitioned into training, validation, and test sets. The training set is used for model training, the validation set for parameter and threshold adjustment, and the test set for performance and generalization evaluation to identify anomalies in the display device circuitry. The deep learning model is YOLOv5s. In terms of software implementation, first, the dataset configuration file and the YOLOv5s model configuration file are modified. Second, the pre-trained weight file is placed in the specified weight path, and the parameters of the weight path are updated accordingly. Then, the parameters and file paths of the modified model YOLOv5_panel.yaml and the dataset panel.yaml are updated. Finally, the YOLOv5s model is trained on the configured system. In this embodiment, the pre-trained weight file is used to store relevant parameter values ​​in the YOLOv5s model.

[0045] After training in S4, the YOLOv5s model generates an accuracy representation to evaluate its accuracy. Accuracy represents the proportion of correctly identified features. Following training, anomaly detection is performed on the display panel using ultrasonic circuit detection. The model identifies circuit defects in the screen's detection area based on these defects, and the ultrasonic echo signal parameters of the abnormal area are statistically analyzed to determine if they meet the preset detection accuracy. If so,... Proceed to step S41: Input the ultrasonic echo signal from the abnormal area into the display device circuit judgment model to judge the defect; save the settings. If not, proceed to step S42: Add a preset amount of noise data to the circuit defect training set and then re-enter step S4 for detection until the preset detection accuracy is reached. In this embodiment, the preset amount of noise data is 1 dB of noise data. The noise data can be the noise from the production line, the typical noise of the equipment itself, or randomly selected operational noise. If the preset detection accuracy requirement is not met, another 1 dB of the same noise data can be added, or another 1 dB of other types of noise data can be added. For example, the first addition is the noise from the production line, and the second addition can be the typical noise of the equipment itself. The specific type of noise can be randomly set.

[0046] S5 uses the display device circuit defect judgment model to determine whether the first detection point in the area to be detected is a defect. If it is, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to judge the defect in the area and perform precision detection. If not, the detection of the first detection point in the area ends.

[0047] S6 confirms whether all areas to be inspected have been measured. If yes, end; otherwise, move the mobile platform to the next to be inspected point to judge defects until all inspection points have been inspected.

[0048] like Figure 3 As shown, a display device circuit judgment model for ultrasonic testing of display devices has been successfully established. When the product under test is detected as defective, the ultrasonic echo signal of the defect is input into the display device circuit judgment model to determine the defect area. If the defect type cannot be determined, the accuracy is then determined. This achieves the effect of simultaneous inspection and measurement.

[0049] The above method establishes a dataset of normal ultrasonic echo signals and defective ultrasonic waves at corresponding coordinate positions of display panels of the same model. It compares the echo signals at defective locations with the normal echo signals to identify circuit abnormalities from abnormal echo detection signals. For abnormal echo signals, corresponding labels are assigned, and an abnormal echo signal dataset is established. Noise is added to the abnormal echo signal dataset to enhance the abnormal echo signals, improving the generalization ability of the dataset and the model. Finally, a deep learning model is used to train the abnormal echo signal dataset, resulting in a circuit defect judgment model for display devices. This model is then used to perform ultrasonic circuit detection on abnormal display panels to confirm whether the abnormal echo signals meet the defined detection accuracy. If the detection accuracy is met, the abnormal ultrasonic echo signals are input into the display device circuit. The defect judgment model, without artificially adding noise, continues to train the model to make the defect ultrasonic dataset more closely resemble the normal ultrasonic dataset, thereby achieving the preset detection accuracy and improving detection precision. Then, the display device circuit defect judgment model obtained from the training set is used to determine whether the first detection point in the area to be inspected is defective. If so, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to determine the defect in the area. If not, the detection of the first detection point ends. It is then confirmed whether all areas to be inspected have been measured. If so, the process ends; otherwise, the moving platform moves to the next to be inspected point to judge defects until all detection points have been inspected. Thus, all inspections can be performed using only the obtained display device circuit defect judgment model, resulting in high detection efficiency.

Claims

1. An ultrasonic circuit testing method for display devices, characterized in that: Includes the following steps: S1 uses an ultrasonic generator to emit ultrasonic waves and record the echo signals of the normal display panel to establish a normal echo signal dataset. S2 analyzes defective display panel samples by using an ultrasonic generator to emit ultrasonic waves and record the echo signals corresponding to the defect locations, forming a defect signal dataset. The received echo signals are converted into electrical signals using an ultrasonic transducer. The echo signal dataset is annotated with the first parameter of the ultrasonic signal using Praat software to form a normal echo signal dataset and a defect signal dataset. A dataset of normal ultrasonic echo signals and defect ultrasonic waves at corresponding coordinate positions of the same model of display panel are established. S3 compares the echo signal at the defect location with the normal echo signal at the same location, and marks the defect type for the defect analysis sample, marking the ultrasonic signal segment of the defect area. The markings are glass crack, micro crack, and circuit break. Based on the echo signal corresponding to the defect location obtained from the defect display panel detection analysis sample, a circuit defect dataset is obtained. The circuit defect dataset is preprocessed and noise data is added. A deep learning model is used to train the circuit defect training set to obtain a circuit defect judgment model for display devices. S4 performs ultrasonic circuit testing on the abnormal display panel, and uses a circuit defect judgment model for display devices to detect circuit abnormalities in the area to be tested on the abnormal screen. It also calculates whether the ultrasonic echo signal parameters of the abnormal area meet the preset detection accuracy. If so, Proceed to step S41: Input the ultrasonic echo signal of the abnormal area into the display device circuit judgment model to judge the defect; if not, proceed to step S42: Add a preset amount of noise data to the circuit defect training set and then re-enter step S4 to perform detection until the preset detection accuracy is reached. S5 uses the display device circuit defect judgment model to determine whether the first detection point in the area to be detected is a defect. If it is, the ultrasonic echo signal of the first detection point in the abnormal area is input into the display device circuit judgment model to judge the area defect. If not, the detection of the first detection point in the area to be detected ends. S6 confirms whether all areas to be inspected have been measured. If yes, end; otherwise, move the mobile platform to the next to be inspected point to judge defects until all inspection points have been inspected.

2. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: In step S1, the normal display panel is moved above a designated position on the normal display panel by a moving platform, and the movement of the moving platform is controlled by the rotation of a control motor.

3. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: In step S2, the defect display panel is inspected and analyzed. The sample is moved above the defect location by a moving platform, and the movement of the moving platform is controlled by the rotation of a control motor.

4. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: Labeling defect types for defect analysis samples includes: analyzing the first parameter of the collected echo signal, comparing the differences in the first parameter, and using software to label the different first parameters in the defect sample.

5. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: The first parameters include the amplitude and frequency of the ultrasound wave.

6. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: Adding noise data involves adding artificial noise to a portion of the randomly selected line defect dataset. This artificial noise is the noise generated during production line operation, or the typical noise inherent in the equipment itself, or randomly selected operational noise.

7. The ultrasonic circuit testing method for display devices according to claim 1, characterized in that: The deep learning model is the YOLOv5s model.

8. The ultrasonic circuit testing method for display devices according to claim 7, characterized in that: The YOLOv5s model requires modification of the dataset configuration file and the YOLOv5s model configuration file. The pre-trained weight file is placed in the specified path of the weights, and the parameters of the weight path are updated accordingly. The parameters and file paths of the modified model YOLOv5_panel.yaml and the dataset panel.yaml are updated. Finally, the YOLOv5s model is trained in the system.

Images