OQC shipment inspection intelligent monitoring system based on AI

Abstract

The invention discloses an OQC shipment inspection intelligent monitoring system based on AI, which belongs to the technical field of shipment inspection and is mainly composed of an intelligent image acquisition system, an AI defect detection system, a size measurement system, a data management and analysis system and a human-computer interaction interface. The intelligent image acquisition system is provided with a high-resolution high-definition camera and an intelligent light source controller, so that high-quality images are obtained, and complex component structures and fine surface features are clearly presented. The AI defect detection system uses a deep learning technology, especially a convolutional neural network model trained by mass data, to accurately and efficiently identify various defects. The dimension measurement system accurately measures the dimension of the component by means of a precision algorithm and a calculation module, and quickly judges whether the component is qualified according to the standard, thereby improving the detection reliability. And the data management and analysis system mines a quality rule through statistical analysis and data mining technologies, and supports decision making. The man-machine interaction interface is simple and visual in design, convenient to operate and clear in result display.

Description

Technical Field

[0001] The present invention relates to the technical field of shipment inspection, and in particular to an AI-based OQC shipment inspection intelligent monitoring system. Background Art

[0002] The OQC shipment inspection system is a key quality control link before the product is delivered to the customer. It covers everything from formulating strict and clear inspection standards, to planning scientific and efficient inspection processes, to using advanced and precise inspection equipment and tools to conduct comprehensive product testing. At the same time, it accurately records and manages the inspection process and results, and ensures that inspectors receive professional training and have the corresponding qualifications, so that they can perform inspection tasks in strict accordance with standards and processes, ensuring that the quality of shipped products fully meets customer requirements and relevant standards.

[0003] As important parts of the engine, cylinder head guards, intake manifolds and other parts have strict quality requirements, and accordingly, the requirements for shipment inspection also need to be improved. However, there are still some problems in the existing OQC shipment inspection process and operation: First, in terms of inspection standards: the current inspection standards are sometimes not clear and detailed enough, and there is a lack of exact judgment criteria for some situations that are on the verge of passing or failing. This can easily lead to differences and inconsistencies in the judgment of specific situations in actual inspection operations, affecting the accuracy and fairness of the inspection results.

[0004] Secondly, in terms of inspection process and equipment: the inspection process may not be simple, efficient or reasonable, which will not only delay the inspection time and increase production costs, but also affect the inspection results. At the same time, some inspection equipment may be unstable due to insufficient precision, aging or improper maintenance, and thus cannot provide accurate and reliable inspection results.

[0005] Finally, in terms of data management, personnel training and information sharing: in terms of data recording and management, they may still rely on traditional manual recording methods, which are prone to errors and are not conducive to data storage, analysis and traceability. The training of inspectors may be incomplete, in-depth or ineffective, which may cause mistakes in the actual work of inspectors. In addition, in terms of information sharing, there is a lack of effective communication and data sharing mechanisms with upstream and downstream supply chains. When quality problems arise, it is difficult to quickly and accurately locate the source of the problem and solve it. Summary of the invention

[0006] The main purpose of the present invention is to provide an AI-based OQC shipment inspection intelligent monitoring system that can effectively solve the problems mentioned in the background technology.

[0007] To achieve the above object, the technical solution adopted by the present invention is: An AI-based OQC shipment inspection intelligent monitoring system includes an intelligent image acquisition system, an AI defect detection system, a dimension measurement system, a data management and analysis system, and a human-computer interaction interface, wherein: The intelligent image acquisition system includes: a high-definition camera, which is responsible for capturing images of parts and components and has high resolution and fast shooting capabilities; an intelligent light source controller, which automatically adjusts the brightness and color of the light source according to ambient lighting conditions; and an image acquisition card, which transmits the image data collected by the camera to the AI ​​defect detection system for processing.

[0008] The AI ​​defect detection system includes: an image preprocessing module, which performs denoising and enhancement preprocessing on the collected images; a deep learning model, which uses a convolutional neural network deep learning algorithm to train a defect detection model; a defect recognition module: which uses the trained model to analyze the image and identify the type, location and size of the defect; and a result marking and output module: which marks the defect detection results on the image and outputs relevant information.

[0009] The dimension measurement system includes: an image processing module, which performs grayscale, enhancement, edge detection, and contour extraction processing on the component image to obtain the feature information required for dimension measurement; a dimension calculation module, which calculates the dimension parameters of the component according to the image processing result; a standard comparison module, which compares the measured dimension with the preset dimension standard to determine whether the dimension is qualified; and an error analysis module, which performs error analysis on the dimension measurement result to evaluate the measurement accuracy.

[0010] The data management and analysis system includes: a database that stores all data during the inspection process; a data acquisition module that is responsible for collecting data from various systems and organizing and storing it; a data analysis module that uses statistical analysis and data mining techniques to analyze data and explore trends and patterns of quality issues; and a report generation module that generates quality statistical reports based on the analysis results.

[0011] The human-computer interaction interface includes: an operation interface, which provides an interface for inspectors to perform inspection operations; a result display interface, which displays the inspection results; a parameter setting interface, which allows inspectors to set inspection parameters and standards; a data upload interface, which supports real-time uploading and sharing of inspection data; and an alarm prompt interface, which issues an alarm prompt when unqualified products are detected to remind inspectors to handle them.

[0012] Preferably, the specific processing method of the image preprocessing module in the AI ​​defect detection system includes: Image denoising: Use the median filter algorithm to remove noise from the image; the algorithm steps of the median filter are as follows: Choose an odd-sized window and slide it across the image.

[0013] For each pixel in the window, replace its grayscale value with the median grayscale value of all pixels in the window.

[0014] Image enhancement: Use the histogram equalization algorithm to improve the quality and contrast of the image; the algorithm steps are as follows: The grayscale histogram of the statistical image is the frequency of occurrence of each gray level.

[0015] Calculate the cumulative histogram, which is the sum of the frequencies of all gray levels before each gray level.

[0016] The grayscale of the original image is mapped by the cumulative histogram to obtain the enhanced image.

[0017] Preferably, the steps for establishing and training the deep learning model are as follows: Data collection: Extensive collection of component images covering both normal and various defects through multiple channels.

[0018] Data annotation: Inspectors accurately annotate the defect type, location, and scope in the image.

[0019] Data partitioning: Divide the data into training set, validation set, and test set in proportion.

[0020] Data preprocessing: uniformly process images, including cropping, normalization, and optional data augmentation.

[0021] Model construction: Select the ResNet architecture and define the structure of each layer, including convolutional layer and pooling layer.

[0022] Model initialization: Initialize with pre-trained parameters and randomly initialize specific layers.

[0023] Training parameter settings: set the learning rate, optimizer and loss function.

[0024] Model training: batch input data, calculate loss, and back-propagate to update parameters.

[0025] Model validation: Use the validation set to evaluate the model and calculate the accuracy.

[0026] Model adjustment: adjust hyperparameters and model structure based on the validation results.

[0027] Model testing: Use the test set to test the model's generalization ability.

[0028] Model saving: Save the model parameters with satisfactory performance for actual testing.

[0029] Preferably, the specific analysis and identification method of the defect identification module includes: Image input: Input the image of the component to be inspected into the system.

[0030] Feature extraction: The convolutional layers and pooling layers in the model extract features from the input image.

[0031] The convolution operation formula is:

[0032] in, The output feature map is The value at The input image is The value at is the convolution kernel in The value at and is the height and width of the convolution kernel, and It is the index used to traverse the elements in the convolution kernel.

[0033] The pooling operation adopts average pooling, and the calculation formula is:

[0034] in, Indicates that the output feature map after the pooling operation is The value of the position, and Respectively represent the height and width of the pooling window, and Respectively represent the step size of the pooling operation in the horizontal and vertical directions, Indicates that the input feature map is The value at the position, and is the index used to traverse the elements within the pooling window.

[0035] Feature propagation: The extracted features are transmitted and integrated layer by layer in the network.

[0036] Classification judgment: The fully connected layer performs classification judgment based on the transmitted features to determine whether there are defects and the type of defects.

[0037] Defect localization: Analyze the intermediate layer feature map to determine the location of the defect in the image.

[0038] Result output: Output the final defect identification results, including defect type and location information.

[0039] Preferably, the specific process of the result marking and output module is as follows: Result acquisition: Receive the defect detection results from the defect recognition module.

[0040] Image marking: Based on the acquired defect location information, the defect area is clearly marked on the original image; different types of defects are distinguished by different marking styles.

[0041] Information Integration: Associate and integrate defect type, severity details with image tags to ensure information consistency and completeness.

[0042] Result output: Output the labeled image and related information.

[0043] Data storage: Store the test results and related image data in the database for subsequent query, statistics and quality traceability.

[0044] Preferably, the specific measurement method in the size measurement system includes four aspects: The specific process of the image processing module is as follows: Image grayscale: Convert a color image into a grayscale image to simplify subsequent processing; convert the red, green, and blue channel values ​​of a color image into a single grayscale value through a weighted average method.

[0045] Image enhancement: The histogram equalization method is used to enhance the contrast of the image; first, the number of pixels at each gray level in the grayscale image is counted to obtain a histogram; then the cumulative histogram is calculated; finally, grayscale mapping is performed based on the cumulative histogram to obtain the enhanced image.

[0046] Image smoothing: Use Gaussian filtering to remove noise, construct a Gaussian kernel, and perform convolution operations on the image to achieve smoothing.

[0047] Edge detection: Using the Canny edge detection algorithm, first perform Gaussian smoothing on the smoothed image again, then calculate the gradient amplitude and direction, and then perform non-maximum suppression on the gradient amplitude, retaining only the local maximum value, and finally use the dual threshold algorithm to detect and connect edges. Set high and low thresholds, and determine pixels with gradient amplitudes greater than the high threshold as edges, and exclude pixels with gradient amplitudes less than the low threshold. For pixels between the two, determine whether they are edges based on their connection with strong edge pixels.

[0048] Contour extraction: Based on the edge detection results, use the contour tracking algorithm to start from an edge point and track along the adjacent edge points in a clockwise or counterclockwise direction until returning to the starting point to form a closed contour. During the tracking process, the pixel coordinates on the contour are recorded.

[0049] The specific process of the size calculation module is as follows: Length and width calculation: For rectangular parts, find the maximum and minimum coordinate values ​​of the contour, length = maximum coordinate value of the X axis - minimum coordinate value of the X axis, width = maximum coordinate value of the Y axis - minimum coordinate value of the Y axis.

[0050] Height calculation: If it is an image of a three-dimensional object, the depth information is obtained through stereo vision or structured light methods to calculate the height.

[0051] Aperture calculation: For a circular aperture, find its contour and calculate the average distance from each point on the contour to the center point as the aperture radius. Aperture = 2 × radius.

[0052] The specific process of the standard comparison module is as follows: Get preset standard size: Read the preset component size standard from the database or configuration file.

[0053] Compare measured dimensions to standard dimensions: For each size parameter, calculate the difference between the measured value and the standard value.

[0054] If the difference is within the allowable tolerance range, it is judged as qualified; otherwise, it is judged as unqualified; The specific process of the error analysis module is as follows: Calculate the error: Error = measured value - true value.

[0055] Analyze the sources of errors: consider the precision of the image acquisition equipment, the accuracy of the image processing algorithm, and the impact of environmental factors on the errors.

[0056] Evaluate measurement accuracy: Use the root mean square error (RMSE) to evaluate measurement accuracy. The formula for RMSE is:

[0057] in, is the true value, is the measured value, is the number of measurements.

[0058] Preferably, the data acquisition module obtains data from each system or module regularly in accordance with the interface specifications between the modules, performs preliminary format verification and integrity check on the acquired data, cleans and converts the collected data to conform to the storage format of the database, and finally inserts the sorted data into the corresponding data table of the database.

[0059] Preferably, the specific analysis method steps of the data analysis module include: Data extraction steps: Data extraction must first clarify the analysis goals, and then use structured query language (SQL) based on these goals to accurately extract the required data from the database.

[0060] Data preprocessing: Data preprocessing includes checking the integrity of the data and handling missing values. For a small number of missing values, the mean, median or mode is used to fill them. For columns with a large number of missing values, their importance is re-evaluated or directly discarded. At the same time, outliers are identified and processed. Data that obviously deviates from the normal range is marked as outliers through statistical methods or based on business logic judgment. Outliers can be deleted, corrected or analyzed separately. In addition, the data needs to be standardized or normalized to make data of different magnitudes and units comparable.

[0061] Statistical analysis: Statistical analysis covers the calculation of basic statistics, such as mean, median, and mode to reflect the central tendency of the data, standard deviation and variance to measure the degree of dispersion of the data, and quartiles to describe the distribution range of the data; it also includes hypothesis testing, such as testing whether there is a significant difference in the mean size of products in different batches, and drawing frequency distribution histograms and box plots to intuitively present the distribution characteristics of the data.

[0062] Data mining: Data mining includes association rule mining, which aims to discover the association between different attributes; it also includes cluster analysis, which aggregates data points with similar characteristics into different clusters to identify potential quality patterns; it also involves classification and prediction, using decision trees and support vector machine algorithms to classify new samples based on existing data or predict whether there are defects.

[0063] Visualization: First, choose the appropriate visualization tools and chart types. Use line charts for time series data, bar charts or pie charts to compare different categories of data, and histograms or kernel density estimation charts to show the distribution of data. Then design a clear chart layout to make the chart easy to understand. Then combine and display multiple charts in a linked manner.

[0064] Preferably, the report generation module refers to designing a report template according to the needs of quality statistics, filling the data of the analysis results according to the format of the template, generating a report file, and distributing the generated report to relevant quality management personnel and decision makers via email or internal system.

[0065] Compared with the prior art, the present invention has the following beneficial effects: 1. Use AI technology to replace traditional manual visual inspection of defects. AI technology can quickly and accurately identify various defects, greatly improve inspection efficiency, and avoid the subjectivity of manual inspection and missed inspections caused by fatigue, ensuring the accuracy and reliability of inspection results.

[0066] 2. Use image processing technology to replace manual measuring tools to measure dimensions. It can achieve fast and accurate dimension measurement, automatically determine whether the dimensions are qualified and give prompts, significantly improve the accuracy and efficiency of measurement, and reduce the impact of human errors on measurement results.

[0067] 3. Build an intelligent data management and analysis system to change the previous data processing methods. It can store inspection data in real time and conduct in-depth analysis to timely discover the trends and patterns of quality problems. Through data sharing, it is convenient for managers to grasp the situation and make timely decisions. Use big data and machine learning technology to predict potential quality problems, take preventive measures in advance, and comprehensively improve the quality management level. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the system composition architecture and flow chart of the present invention. DETAILED DESCRIPTION

[0069] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0070] like Figure 1 As shown in the figure, an AI-based OQC shipment inspection intelligent monitoring system is mainly composed of an intelligent image acquisition system, an AI defect detection system, a dimension measurement system, a data management and analysis system, and a human-computer interaction interface. Its workflow is as follows: The intelligent image acquisition system is started, and the high-definition camera, with the cooperation of the intelligent light source, performs all-round image acquisition of engine hood, manifold and other components, and transmits the image data to the computer through the image acquisition card.

[0071] The AI ​​defect detection system starts working. The image preprocessing module performs denoising, enhancement and other preprocessing on the collected images. Then the deep learning model analyzes the images. The defect recognition module identifies the type, location and size of the defects. Finally, the result marking and output module marks the defect detection results on the image and outputs relevant information.

[0072] The image processing module in the dimension measurement system processes the component images output by the AI ​​defect detection system to obtain the feature information required for dimension measurement. The dimension calculation module calculates the dimension parameters of the components. The standard comparison module compares the measured dimensions with the preset standards to determine whether the dimensions are qualified. The error analysis module performs error analysis on the dimension measurement results.

[0073] The data acquisition module of the data management and analysis system collects data from various systems, such as the intelligent image acquisition system, AI defect detection system, and dimension measurement system, including images, measurement results, and inspection reports, and stores them in the database. The data analysis module analyzes the data to explore the trends and patterns of quality issues, and the report generation module generates quality statistical reports based on the analysis results.

[0074] The human-computer interaction interface provides an operation interface for inspectors, through which they can view images, mark defects, measure dimensions, and perform other operations. The result display interface displays the inspection results, including defect information, dimension data, and pass / fail status. The parameter setting interface allows inspectors to set inspection parameters and standards. When unqualified products are detected, the alarm prompt interface issues an alarm prompt to remind inspectors to handle them. Inspection data is uploaded and shared in real time through the data upload interface, which facilitates remote monitoring and decision-making by managers.

[0075] The following introduces the specific implementation method to introduce the system in detail: Intelligent Image Acquisition System

[0076] High-definition cameras are placed at key locations on the production line, with a resolution of up to 8K and a high-speed shooting capability of 120 frames per second, which can accurately capture the subtle details of engine components. For example, the tiny holes on the cylinder head cover and the fine scratches on the surface of the intake manifold can all be captured with clear and detailed images.

[0077] The intelligent light source controller can automatically and sensitively adjust the brightness and color of the light source according to the actual lighting conditions in the factory workshop. In the period of sufficient light, it will appropriately reduce the brightness of the light source to avoid overexposure of the image; in the case of dim light, it will increase the brightness of the light source and adjust it to the appropriate color temperature to ensure that the captured image always maintains high quality.

[0078] The image acquisition card quickly transmits the image data captured by the camera to the AI ​​defect detection system for subsequent processing at an extremely fast data transmission speed, ensuring the real-time and efficiency of the entire process.

[0079] AI Defect Detection System

[0080] The image preprocessing module performs a series of processing operations on the collected images. In the image denoising stage, the median filter algorithm selects a 5x5 window to slide on the image, and uses the median of the pixel grayscale value in the window as the new grayscale value of the center pixel, successfully removing the salt and pepper noise and random noise in the image. For example, for an image of a cylinder head cover with noise, after median filtering, the surface noise is significantly reduced, and potential defects are more clearly displayed.

[0081] The image enhancement part uses a histogram equalization algorithm to significantly increase the contrast of the image. For example, the original low-contrast image of the intake manifold becomes more obvious and easier to identify after processing.

[0082] The establishment and training of the deep learning model went through multiple rigorous steps. During the data collection phase, a wide range of images of automobile engine parts covering normal and various defects (such as cracks, sand holes, injection defects, etc.) were collected from the factory's past production data, industry shared databases, and data obtained with partners, totaling 20,000 images.

[0083] During the data annotation process, experienced and professional inspectors use high-precision annotation tools to accurately mark the defect type, location and range in each image.

[0084] The data is divided into training set, validation set and test set in a ratio of 6:3:1.

[0085] In the process of data preprocessing, the images were uniformly cropped and resized, and grayscale normalized. At the same time, data enhancement operations such as random flipping and scaling were also used to enrich the diversity of the data.

[0086] The ResNet101 architecture was selected for model construction, and structures such as convolutional layers and pooling layers were carefully defined.

[0087] The model initialization utilizes parameters pre-trained on a large-scale image dataset and performs random initialization on specific layers.

[0088] When setting the training parameters, the learning rate was initially set to 0.0005, the Adagrad optimizer was used, and the cross entropy loss function was selected as the loss function.

[0089] During the model training process, 64 images were input in batches each time, and the parameters were updated through back propagation after calculating the loss. After 100 epochs of training, the accuracy on the validation set reached 96%.

[0090] Based on the verification results, the learning rate was dynamically adjusted and two convolutional layers were added. After training again, the accuracy of the model on the verification set was improved to 99%.

[0091] Finally, the generalization ability of the model was tested on the test set, and the accuracy rate reached 98%, meeting the high requirements of actual testing. The model was properly saved for subsequent actual testing.

[0092] The defect recognition module performs in-depth analysis on the input image of the automobile engine component. When the image is input, feature extraction is performed with the help of convolutional layers and pooling layers. In the convolution operation, for example, a 7x7 convolution kernel is convolved with a local area of ​​the input image to effectively extract key features.

[0093] The pooling operation uses maximum pooling, such as a 3x3 pooling window with a step size of 3, to downsample the feature map, thereby reducing the amount of data while highlighting the main features.

[0094] Features are transmitted and integrated layer by layer in the network, and the fully connected layer makes accurate classification judgments based on these features to determine whether there are defects and the type of defects. By deeply analyzing the feature map of the middle layer, the specific location of the defect in the component image can be accurately locked.

[0095] The result marking and output module clearly marks the defect detection results on the image. For example, for the crack defect on the intake manifold, the defect location is marked with a striking blue line and the words "severe crack" are marked next to it. At the same time, detailed information such as defect type and severity is closely integrated with image marking, and the marked image and related information are output, and all data is properly stored in the database.

[0096] Dimensional measurement system

[0097] The image processing module performs fine processing on the images of automobile engine parts. The image grayscale converts the color cylinder head cover image into a grayscale image to simplify the data for subsequent processing.

[0098] Image enhancement uses histogram equalization to significantly enhance the contrast of component details in the image. For example, the contour and size features of the intake manifold mounting hole, which was originally dark, are more clearly discernible after processing.

[0099] Image smoothing uses Gaussian filtering to effectively remove noise. For example, for an image of a noisy cylinder head cover fastening screw hole, the noise is greatly reduced after filtering.

[0100] Edge detection uses the Canny algorithm to accurately detect the edge contours of components. For example, the complex edges of the intake manifold can be accurately detected and extracted.

[0101] Contour extraction uses contour tracking algorithm to obtain complete and continuous component contours.

[0102] In the size calculation module, for the rectangular cylinder head cover, the length and width are accurately calculated by precisely finding the maximum and minimum coordinate values ​​of the contour.

[0103] For the height of the curved part of the intake manifold, the height is calculated by acquiring three-dimensional point cloud data using laser scanning technology.

[0104] For the mounting hole diameter on the cylinder head cover, the average distance from each point on its contour to the center point is calculated as the hole radius.

[0105] The standard comparison module obtains the preset automobile engine component size standards from the database and carefully calculates the difference between the measured value and the standard value. For example, the diameter standard of a mounting hole on the cylinder head shield is 20mm, and the measured value is 19.8mm. The difference is within the allowable tolerance range of 0.2mm and is judged to be qualified.

[0106] The error analysis module accurately calculates the error and deeply analyzes the source of the error. For example, when measuring the length of the intake manifold, the error mainly comes from the accuracy of the image acquisition device and the limitations of the measurement algorithm. The measurement accuracy is evaluated to be 99%, which meets the strict production requirements.

[0107] Data management and analysis system

[0108] The database comprehensively stores all inspection data during the production process of automotive engine parts, including images, measurement results, test reports, equipment parameters and other key information.

[0109] The data acquisition module strictly follows the interface specifications and obtains data from each system every half hour. The acquired data is strictly formatted and checked for integrity, such as carefully checking whether the measured value is within a reasonable range. The collected data is fully cleaned and converted, such as converting the temperature unit from Fahrenheit to Celsius, and then accurately inserted into the corresponding data table in the database.

[0110] The data analysis module first defines the specific analysis objectives, such as analyzing the size deviation distribution trend of intake manifolds in different batches, and uses structured query language (SQL) to accurately extract relevant data from the database.

[0111] In the data preprocessing phase, we carefully check the data integrity and properly handle missing values. For example, for a small amount of missing production time data, we use the median value of adjacent data to fill it. We accurately identify and handle outliers, such as marking dimensional measurements that are significantly deviated from the normal range as outliers and conducting in-depth separate analysis. We standardize the data to ensure that the data is highly comparable.

[0112] Statistical analysis accurately calculates the mean and standard deviation of the dimensions of intake manifolds from different batches. Performs rigorous hypothesis testing to accurately determine whether there are significant differences in the dimensional deviations of different batches. Carefully draws frequency distribution histograms and box plots to intuitively and clearly display the distribution of dimensional deviations.

[0113] Data mining successfully discovered the close relationship between intake manifold defect types and production process parameters through association rule mining. For example, it was revealed that intake manifolds produced at a specific casting temperature are more likely to have porosity defects. Through cluster analysis, intake manifolds with similar defect characteristics were grouped together. A decision tree algorithm was used to predict defects for newly produced parts.

[0114] Visualization Carefully select appropriate visualization tools and chart types, such as using line charts to dynamically display the changing trend of cylinder head cover size deviations in different batches, using bar charts to intuitively compare the defect rates of intake manifolds under different production processes, and using pie charts to clearly display the proportion of various defect types in all defects. Carefully design clear and easy-to-understand chart layouts, and cleverly combine and link multiple charts to facilitate comprehensive analysis from multiple angles.

[0115] The report generation module carefully designs the quality report template for automobile engine parts according to strict quality statistics requirements. The data of the analysis results are accurately filled in according to the template format, and a detailed Excel report file is generated, which is distributed to quality management personnel and decision makers on a daily basis through the internal system.

[0116] Human-computer interaction interface

[0117] The operation interface provides inspectors with convenient and intuitive inspection operation functions. For example, inspectors can easily mark defects on the image through the touch screen, or use specific shortcut keys for quick operation.

[0118] The result display interface presents the inspection results of automobile engine components in a clear and concise manner, such as whether the cylinder head cover has defects, whether the intake manifold size is qualified, and other key information.

[0119] The parameter setting interface allows inspectors to flexibly set inspection parameters and strict standards according to different models of automobile engine components.

[0120] The data upload interface supports real-time upload of inspection data to the factory’s central server, enabling data sharing and collaborative work among multiple departments.

[0121] When the alarm prompt interface detects unqualified parts, it will immediately send out a strong sound and light alarm to remind the inspectors to deal with them quickly.

[0122] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An AI-based OQC shipment inspection intelligent monitoring system, characterized by: It includes intelligent image acquisition system, AI defect detection system, dimension measurement system, data management and analysis system and human-computer interaction interface, among which: The intelligent image acquisition system includes: a high-definition camera, which is responsible for capturing images of parts and components, and has high resolution and fast shooting capabilities; an intelligent light source controller, which automatically adjusts the brightness and color of the light source according to the ambient light conditions; an image acquisition card, which transmits the image data collected by the camera to the AI ​​defect detection system for processing; The AI ​​defect detection system includes: an image preprocessing module, which performs denoising and enhancement preprocessing on the collected images; a deep learning model, which uses a convolutional neural network deep learning algorithm to train a defect detection model; a defect recognition module: which uses the trained model to analyze the image and identify the type, location and size of the defect; a result marking and output module: which marks the defect detection results on the image and outputs relevant information; The dimension measurement system includes: an image processing module, which performs grayscale, enhancement, edge detection, and contour extraction processing on the component image to obtain the feature information required for dimension measurement; a dimension calculation module, which calculates the dimension parameters of the component according to the image processing result; a standard comparison module, which compares the measured dimension with the preset dimension standard to determine whether the dimension is qualified; and an error analysis module, which performs error analysis on the dimension measurement result to evaluate the measurement accuracy. The data management and analysis system includes: a database that stores all data during the inspection process; a data acquisition module that is responsible for collecting data from various systems, sorting and storing it; a data analysis module that uses statistical analysis and data mining techniques to analyze data and mine trends and patterns of quality issues; and a report generation module that generates quality statistical reports based on the analysis results. The human-computer interaction interface includes: an operation interface, which provides an interface for inspectors to perform inspection operations; a result display interface, which displays the inspection results; a parameter setting interface, which allows inspectors to set inspection parameters and standards; a data upload interface, which supports real-time uploading and sharing of inspection data; and an alarm prompt interface, which issues an alarm prompt when unqualified products are detected to remind inspectors to handle them.

2. According to claim 1, an AI-based OQC shipment inspection intelligent monitoring system is characterized by: The specific processing method of the image preprocessing module in the AI ​​defect detection system includes: Image denoising: Use the median filter algorithm to remove noise from the image; the algorithm steps of the median filter are as follows: Choose an odd-sized window and slide it across the image; For each pixel in the window, replace its gray value with the median gray value of all pixels in the window; Image enhancement: Use the histogram equalization algorithm to improve the quality and contrast of the image; the algorithm steps are as follows: Statistical image grayscale histogram, that is, the frequency of each gray level; Calculate the cumulative histogram, which is the sum of the frequencies of all gray levels before each gray level; The grayscale of the original image is mapped by the cumulative histogram to obtain the enhanced image.

3. According to claim 2, an AI-based OQC shipment inspection intelligent monitoring system is characterized by: The steps for establishing and training the deep learning model are as follows: Data collection: widely collect images of parts with normal and various defects through multiple channels; Data annotation: Inspectors accurately annotate the defect type, location, and scope in the image; Data partitioning: Divide the data into training set, validation set and test set in proportion; Data preprocessing: uniformly process images, including cropping, normalization, and optional data augmentation; Model construction: Select the ResNet architecture and define the structure of each layer, including convolutional layer and pooling layer; Model initialization: Initialize with pre-trained parameters and randomly initialize specific layers; Training parameter settings: setting learning rate, optimizer and loss function; Model training: batch input data, calculate loss, and back-propagate to update parameters; Model validation: Use the validation set to evaluate the model and calculate the accuracy; Model adjustment: adjust hyperparameters and model structure according to the verification results; Model testing: Use the test set to test the generalization ability of the model; Model saving: Save the model parameters with satisfactory performance for actual testing.

4. According to claim 3, an AI-based OQC shipment inspection intelligent monitoring system is characterized by: The specific analysis and identification method of the defect identification module includes: Image input: input the image of the component to be inspected into the system; Feature extraction: The convolutional layers and pooling layers in the model extract features from the input image; The convolution operation formula is: in, The output feature map is The value at The input image is The value at is the convolution kernel in The value at and is the height and width of the convolution kernel, and It is the index used to traverse the elements in the convolution kernel; The pooling operation adopts average pooling, and the calculation formula is: in, Indicates that the output feature map after the pooling operation is The value of the position, and Respectively represent the height and width of the pooling window, and Respectively represent the step size of the pooling operation in the horizontal and vertical directions, Indicates that the input feature map is The value at the position, and is the index used to traverse the elements in the pooling window; Feature propagation: The extracted features are transmitted and integrated layer by layer in the network; Classification judgment: The fully connected layer classifies and judges based on the transmitted features to determine whether there are defects and the defect type; Defect location: Analyze the feature map of the middle layer to determine the location of the defect in the image; Result output: Output the final defect identification results, including defect type and location information.

5. According to claim 4, an AI-based OQC shipment inspection intelligent monitoring system is characterized by: The specific process of the result marking and output module is as follows: Result acquisition: receiving defect detection results from the defect recognition module; Image marking: Based on the acquired defect location information, clearly mark the defect area on the original image; use different marking styles to distinguish different types of defects; Information integration: Associate and integrate defect type, severity details with image tags to ensure consistency and completeness of information; Result output: output the marked image and related information; Data storage: Store the test results and related image data in the database for subsequent query, statistics and quality traceability.

6. The AI-based OQC shipment inspection intelligent monitoring system according to claim 1 is characterized by: The specific measurement method in the dimension measurement system includes four aspects: The specific process of the image processing module is as follows: Image grayscale: Convert color images to grayscale images to simplify subsequent processing; convert the red, green, and blue channel values ​​of color images into a single grayscale value through the weighted average method; Image enhancement: The contrast of the image is enhanced by using the histogram equalization method. First, the number of pixels at each gray level in the grayscale image is counted to obtain a histogram. Then, the cumulative histogram is calculated. Finally, grayscale mapping is performed according to the cumulative histogram to obtain the enhanced image; Image smoothing: Use Gaussian filtering to remove noise, construct a Gaussian kernel, and perform convolution operations on the image to achieve smoothing; Edge detection: Using the Canny edge detection algorithm, first perform Gaussian smoothing on the smoothed image again, then calculate the gradient amplitude and direction, then perform non-maximum suppression on the gradient amplitude, retaining only the local maximum value, and finally use the dual threshold algorithm to detect and connect edges. Set high and low thresholds, and determine pixels with gradient amplitudes greater than the high threshold as edges, and exclude pixels with gradient amplitudes less than the low threshold. For pixels between the two, determine whether they are edges based on their connection with strong edge pixels; Contour extraction: Based on the edge detection results, use the contour tracking algorithm to start from an edge point and track along the adjacent edge points in a clockwise or counterclockwise direction until returning to the starting point to form a closed contour. During the tracking process, the pixel coordinates on the contour are recorded; The specific process of the size calculation module is as follows: Length and width calculation: For rectangular parts, find the maximum and minimum coordinate values ​​of the contour, length = maximum coordinate value of the X axis - minimum coordinate value of the X axis, width = maximum coordinate value of the Y axis - minimum coordinate value of the Y axis; Height calculation: If it is an image of a three-dimensional object, the depth information is obtained through stereo vision or structured light methods to calculate the height; Aperture calculation: For a circular aperture, find its contour and calculate the average distance from each point on the contour to the center point as the aperture radius. Aperture = 2 × radius. The specific process of the standard comparison module is as follows: Get preset standard size: read the preset component size standard from the database or configuration file; Compare measured dimensions with standard dimensions: For each dimension parameter, calculate the difference between the measured value and the standard value; If the difference is within the allowable tolerance range, it is judged as qualified; otherwise, it is judged as unqualified; The specific process of the error analysis module is as follows: Calculate the error: Error = measured value - true value; Analyze the sources of errors: consider the precision of the image acquisition equipment, the accuracy of the image processing algorithm, and the impact of environmental factors on the errors; Evaluate measurement accuracy: Use the root mean square error (RMSE) to evaluate measurement accuracy. The formula for RMSE is: in, is the true value, is the measured value, is the number of measurements.

7. The AI-based OQC shipment inspection intelligent monitoring system according to claim 1 is characterized by: The data acquisition module obtains data from each system or module regularly according to the interface specifications between the modules, performs preliminary format verification and integrity check on the acquired data, cleans and converts the collected data to make it conform to the storage format of the database, and finally inserts the sorted data into the corresponding data table of the database.

8. The AI-based OQC shipment inspection intelligent monitoring system according to claim 1 is characterized by: The specific analysis method steps of the data analysis module include: Data extraction steps: Data extraction first requires clear analysis objectives, and then based on these objectives, structured query language (SQL) is used to accurately extract the required data from the database; Data preprocessing: Data preprocessing includes checking the integrity of the data and handling missing values. For a small number of missing values, the mean, median or mode is used to fill them. For columns with a large number of missing values, their importance is re-evaluated or directly discarded. At the same time, outliers are identified and processed. Data that obviously deviates from the normal range is marked as outliers through statistical methods or based on business logic judgment. Outliers can be deleted, corrected or analyzed separately. In addition, the data needs to be standardized or normalized to make data of different magnitudes and units comparable. Statistical analysis: Statistical analysis includes calculating basic statistics, such as mean, median, mode to reflect the central tendency of data, standard deviation and variance to measure the dispersion of data, and quartiles to describe the distribution range of data. It also includes hypothesis testing, such as testing whether there is a significant difference in the mean size of products in different batches, and drawing frequency distribution histograms and box plots to intuitively present the distribution characteristics of data. Data mining: Data mining includes association rule mining, which aims to discover the association between different attributes; it also includes cluster analysis, which aggregates data points with similar characteristics into different clusters to identify potential quality patterns; it also involves classification and prediction, using decision trees and support vector machine algorithms to classify new samples based on existing data or predict whether there are defects; Visualization: First, choose the appropriate visualization tools and chart types. Use line charts for time series data, bar charts or pie charts to compare different categories of data, and histograms or kernel density estimation charts to show the distribution of data. Then design a clear chart layout to make the chart easy to understand. Then combine and display multiple charts in a linked manner.

9. The AI-based OQC shipment inspection intelligent monitoring system according to claim 8 is characterized by: The report generation module is to design a report template according to the needs of quality statistics, fill the data of the analysis results according to the format of the template, generate a report file, and distribute the generated report to relevant quality management personnel and decision makers via email or internal system.

Images