Full-automatic steel plate internal defect online detection system based on ultrasonic phased array

Abstract

The invention discloses a full-automatic steel plate internal defect on-line detection system based on an ultrasonic phased array, and relates to the technical field of nondestructive testing, and the system comprises an intelligent analysis module which is in communication connection with an on-line detection module and is used for operating an artificial intelligence rating model to carry out processing and preliminary evaluation on original ultrasonic detection data, outputting an initial evaluation result containing the defect space position; and the high-precision verification module is in communication connection with the intelligent analysis module, and is used for intercepting a corresponding sample steel plate according to the defect space position in the preliminary evaluation result, and verifying and measuring the sample steel plate through a high-precision detection means to generate defect true value data. According to the invention, the double quality control modules are combined with the look-ahead control and the backtracking control, so that the subsequent steel plate detection quality is guaranteed, the historical detection data deviation is corrected, the whole-cycle quality control is realized, the missing detection rate and the false detection rate are effectively reduced, and the quality stability of steel plate products is improved.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to a fully automated online detection system for internal defects in steel plates based on an ultrasonic phased array. Background Technology

[0002] As the steel industry transforms towards high-end and intelligent manufacturing, key sectors have stringent requirements for the internal quality of steel plates. Defects such as porosity, inclusions, and cracks within the steel plate directly affect the structural safety and service life of products, and can even lead to major safety accidents. Therefore, achieving real-time, accurate, and fully automated online detection of internal defects during steel plate rolling has become a core requirement for ensuring product quality and improving production efficiency. Traditional offline detection requires sampling after production is completed, which not only fails to provide real-time feedback on production quality and timely adjustments to process parameters, but also easily leads to missed defects due to insufficient sampling representativeness, making it difficult to adapt to the continuous high-speed production rhythm of modern steel plate rolling lines.

[0003] From the perspective of technological development, the detection of internal defects in steel plates has gradually evolved towards automation and intelligence. In the early stages, offline detection using handheld ultrasonic probes was the main method, resulting in low detection efficiency, high labor intensity, and results that were highly dependent on the operator's experience and had poor consistency. Subsequently, traditional ultrasonic testing systems were applied to online scenarios, using single-element or linear array probes for conventional scanning. Although this could initially achieve continuous detection, the coverage was limited, the beam focusing flexibility was insufficient, and it was difficult to accurately locate internal defects in moving steel plates. In recent years, ultrasonic phased array technology has significantly improved detection coverage and sensitivity with its advantages of multi-element controllable focusing and flexible beam deflection. Artificial intelligence technology has also begun to be combined with ultrasonic testing to achieve automatic defect identification and rating. However, existing technologies mostly focus on single-stage functions and have not yet formed a complete closed loop from raw data acquisition and intelligent initial assessment to high-precision verification and model optimization. The stability of online detection and the accuracy of defect location still need to be improved.

[0004] The current technology system still faces several challenges: First, online detection stability is insufficient. The movement of the steel plate can easily cause the ultrasonic probe scanning area to deviate from the target. Simultaneously, the coupling medium is affected by temperature and pressure fluctuations, resulting in uneven water film thickness or breakage, leading to decreased ultrasonic wave transmission efficiency and distortion of the original detection data. Second, the intelligent analysis model has poor adaptability. Existing models are trained on fixed datasets and lack dynamic optimization mechanisms. When initial assessment results deviate from actual defects in type, size, or rating, they cannot adjust autonomously, leading to a gradual decrease in detection accuracy over long-term use. Third, the high-precision verification mechanism is inflexible, often triggered by fixed time periods or the number of tests, rather than dynamically based on the initial defect assessment confidence level, resulting in delayed verification of high-risk defects. Fourth, quality control lacks systematicity. Optimized models can only be applied to subsequent inspections and cannot retrospectively correct historical inspection data, making it difficult to form a quality control system covering the entire production lifecycle. These problems collectively restrict detection accuracy and reliability, failing to meet the quality control requirements of high-end steel products.

[0005] Therefore, it is essential to invent a fully automated online detection system for internal defects in steel plates based on ultrasonic phased arrays to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a fully automated online detection system for internal defects in steel plates based on ultrasonic phased array, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a fully automated online detection system for internal defects in steel plates based on ultrasonic phased array, comprising:

[0008] The online inspection module is used to perform non-destructive testing on moving steel plates on the steel plate rolling line and generate raw ultrasonic test data;

[0009] The intelligent analysis module is communicatively connected to the online detection module and is used to run an artificial intelligence rating model to process and preliminarily evaluate the original ultrasonic detection data, and output a preliminary evaluation result including the spatial location of the defects.

[0010] The high-precision verification module is connected in communication with the intelligent analysis module. It is used to cut out the corresponding sample steel plate according to the spatial location of the defect in the preliminary evaluation result, and to verify and measure it through high-precision detection methods to generate true defect data.

[0011] The model optimization module is communicatively connected to the high-precision verification module and is used to optimize the artificial intelligence rating model based on the deviation between the defect truth data and the initial evaluation result.

[0012] A dual quality control module, coupled with the intelligent analysis module, is used to deploy the optimized artificial intelligence rating model to the online detection module for subsequent steel plate detection, and to perform retrospective simulation screening and correction of historical detection data.

[0013] Preferably, the online detection module includes:

[0014] The mechanical conveying and centering unit is used to convey steel plates and ensure that they are centered as they pass through the inspection area;

[0015] The coupling unit is used to provide a stable coupling medium between the ultrasonic testing probe and the steel plate surface;

[0016] The core detection unit is used to collect and generate raw ultrasound detection data.

[0017] Preferably, the raw ultrasonic testing data includes full matrix capture (FMC) data and steel plate position encoder information.

[0018] Preferably, the intelligent analysis module includes:

[0019] The defect intelligent identification unit is used to receive the raw ultrasonic test data from the online detection module, process the raw ultrasonic test data to obtain complete ultrasonic test data, input the complete ultrasonic test data into the initial artificial intelligence rating model, and output the physical characteristic information of the defect, including defect type, confidence level, spatial location and predicted size.

[0020] The rating unit is used to output the rating result of the defect based on the physical characteristics of the defect and according to the preset rating criteria, which together constitute the initial rating result.

[0021] Preferably, the high-precision verification module includes:

[0022] The automatic sample cutting unit is used to cut sample steel plates containing specific defects from the production line according to the instructions of the intelligent analysis module.

[0023] The high-precision verification unit is an industrial CT scanning system used to scan the sample steel plate to obtain the defect type, size and defect level, and generate defect true value data.

[0024] Preferably, the triggering condition for the high-precision verification module is at least one of the following:

[0025] The system's continuous production time has reached the preset time cycle;

[0026] The system has cumulatively detected a number of steel plates that have reached a preset threshold.

[0027] The intelligent analysis module's initial assessment confidence level for a certain type of defect is lower than the preset confidence threshold.

[0028] Preferably, the model optimization module includes:

[0029] The deviation calculation unit is used to compare the initial evaluation results of the rating unit with the defect true value data generated by the high-precision verification unit, and calculate the type deviation, size deviation, rating deviation and comprehensive deviation degree.

[0030] The model optimization unit is used to perform backpropagation and parameter fine-tuning on the initial artificial intelligence rating model, using the comprehensive deviation as part of the loss function, to generate an optimized artificial intelligence rating model.

[0031] Preferably, the deviation calculation unit is configured as follows:

[0032] The type deviation is calculated using the cross-entropy loss function;

[0033] The dimensional deviation is calculated using the mean square error.

[0034] The rating deviation is calculated using the grade difference;

[0035] The overall deviation is a function f(δ) type ,δ size ,δ level The calculation result of ), where δ type For type bias, δ size For dimensional deviation, δ level This indicates a rating deviation.

[0036] Preferably, the dual quality control module includes:

[0037] The forward-looking control unit is used to deploy the optimized artificial intelligence rating model to the intelligent analysis module to detect the steel plates that subsequently pass through the online detection module;

[0038] The retrospective control unit is used to call the optimized artificial intelligence rating model to recalculate and rate the stored historical detection data in order to identify and correct potential false detection events in the past and generate a correction report.

[0039] The technical effects and advantages of this invention are as follows:

[0040] 1. This invention achieves stable, accurate, and non-destructive testing of steel plates in motion by coordinating the mechanical transmission of the online detection module with the centering unit, coupling unit, and core detection unit. The generated raw ultrasonic test data has high synchronicity and integrity, laying the foundation for subsequent defect identification.

[0041] 2. This invention achieves automatic defect identification and preliminary assessment by combining an intelligent analysis module with an artificial intelligence rating model. Compared with traditional manual analysis, it significantly improves detection efficiency and consistency of preliminary assessment. At the same time, the output defect spatial location provides accurate positioning for high-precision verification.

[0042] 3. This invention uses a high-precision verification module to obtain true defect data through industrial CT scanning, and combines it with a multi-condition triggering mechanism to ensure the timeliness and accuracy of verification, providing reliable benchmark data for model optimization;

[0043] 4. This invention uses a model optimization module to fine-tune model parameters based on deviation calculation, thereby continuously improving the accuracy of the artificial intelligence rating model and solving the problem that traditional detection models are easily affected by the environment and steel plate material, leading to a decrease in accuracy.

[0044] 5. This invention combines forward-looking and retrospective control through a dual quality control module, which not only ensures the quality of subsequent steel plate testing but also corrects deviations in historical testing data, achieving full-cycle quality control, effectively reducing the rate of missed and false detections, and improving the quality stability of steel plate products. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the system architecture of the present invention.

[0046] Figure 2 This is a diagram showing the composition of the online detection module of the present invention.

[0047] Figure 3 This is a diagram showing the composition of the intelligent analysis module of the present invention.

[0048] Figure 4 This is a diagram showing the trigger conditions for the high-precision verification module of the present invention.

[0049] Figure 5 This is a diagram illustrating the working mechanism of the model optimization module of the present invention.

[0050] Figure 6 This is a flowchart of the dual quality control module of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] This invention provides, for example Figure 1The fully automated online inspection system for internal defects in steel plates based on ultrasonic phased array shown includes an online inspection module, an intelligent analysis module, a high-precision verification module, a model optimization module, and a dual quality control module.

[0053] The online inspection module is used to perform non-destructive testing on moving steel plates on the steel plate rolling line and generate raw ultrasonic test data;

[0054] Furthermore, in the above technical solution, the online detection module is as follows: Figure 2 The following are included:

[0055] The mechanical conveying and centering unit is used to convey steel plates and ensure that they are centered as they pass through the inspection area;

[0056] The coupling unit is used to provide a stable coupling medium between the ultrasonic testing probe and the steel plate surface;

[0057] The core detection unit is used to collect and generate raw ultrasound detection data.

[0058] Furthermore, in the above technical solution, the original ultrasonic detection data includes full matrix capture (FMC) data and steel plate position encoder information.

[0059] It is important to understand that the mechanical conveying and centering unit mainly includes a roller conveyor drive motor, conveyor rollers, and a photoelectric centering sensing system. The rolled steel plate is carried and transported by the conveyor rollers. Before entering the inspection area, photoelectric sensor arrays located on both sides of the roller conveyor detect the edge position of the steel plate in real time and transmit the position signal to the PLC. The PLC calculates the position deviation and controls the servo motors of the correction rollers on one or both sides to make fine adjustments, ensuring that the deviation between the centerline of the steel plate and the centerline of the probe scan is controlled within ±10mm when the steel plate passes through the core inspection unit. This unit effectively solves the problem of missed detection caused by steel plate deviation.

[0060] The PLC is a programmable logic controller;

[0061] The coupling unit includes a constant temperature water tank, a water pump, a pressure regulating valve, a flow meter, and a spray device installed in front of the probe. The constant temperature water tank controls the water temperature at 40±5℃ to reduce the impact of water temperature fluctuations on the speed of sound. The water pump delivers water to the spray device, and the water pressure is stabilized at 0.2~0.4MPa and the flow rate is stabilized at 3~5L / min through the regulating valve and the flow meter, thereby forming a uniform and stable water film between the probe and the steel plate surface, providing a stable path for ultrasonic wave transmission. A leakage recovery tank is provided behind the spray device for collecting and recycling the coupling water.

[0062] The core detection unit employs multiple linear ultrasonic phased array probes, for example, 64 elements with a 5MHz center frequency, arranged side-by-side to achieve full coverage of the steel plate width. The probe holder incorporates a servo motor, driving the probes to scan along the width of the steel plate. The probes are connected to a high-speed ultrasonic phased array acquisition instrument. Under FPGA control, this acquisition instrument transmits and receives ultrasonic signals in full-matrix acquisition mode. In this mode, the acquisition instrument sequentially excites each element in the probe array as a transmission source and simultaneously acquires the received signals from all elements, thereby obtaining the most complete original A-scan signal group containing all possible sound wave propagation paths. Simultaneously, a high-precision photoelectric encoder, such as the OMRON E6B2-CWZ6C, is coaxially connected to the rollers of the conveyor belt via a coupling. This converts the rollers' rotational motion into high-precision electrical pulse signals. The pulses generated by the encoder are counted in real-time by a high-speed counter and converted into the real-time displacement of the steel plate. Whenever the displacement reaches a preset interval, such as 0.5mm, a synchronous control circuit composed of an FPGA sends a trigger signal to the ultrasonic phased array acquisition instrument, driving it to immediately complete one FMC data acquisition. At the same time, the core detection unit locks the current encoder count value and converts it into the steel plate's position coordinates (X direction), strictly binding it to all ultrasonic data acquired in this session. Finally, this unit outputs a strictly synchronized raw ultrasonic detection data in real-time, containing full-matrix capture FMC data and steel plate position encoder information.

[0063] It should be noted that the FPGA is a field-programmable gate array;

[0064] The intelligent analysis module is communicatively connected to the online detection module and is used to run an artificial intelligence rating model to process and preliminarily evaluate the original ultrasonic detection data, and output a preliminary evaluation result including the spatial location of the defects.

[0065] Furthermore, in the above technical solution, the intelligent analysis module is as follows: Figure 3 The following are included:

[0066] The defect intelligent identification unit is used to receive the raw ultrasonic test data from the online detection module, process the raw ultrasonic test data to obtain complete ultrasonic test data, input the complete ultrasonic test data into the initial artificial intelligence rating model, and output the physical characteristic information of the defect, including defect type, confidence level, spatial location and predicted size.

[0067] The rating unit is used to output the rating result of the defect based on the physical characteristics of the defect and according to the preset rating criteria, which together constitute the initial rating result.

[0068] It should be noted that the complete ultrasound detection data includes full-focus TFM image data and steel plate position encoder information; the full-focus TFM image data is obtained by processing full-matrix capture FMC data through a full-focus imaging algorithm;

[0069] The rating results include Level 1 representing qualified, Level 2 representing pending observation, Level 3 representing rework, and Level 4 representing scrap.

[0070] The initial artificial intelligence rating model is preferably a deep learning-based convolutional neural network model, such as U-Net or YOLO architecture, trained on a large amount of labeled TFM image data of steel plate defects. The model analyzes the input fully focused TFM image data and outputs the physical feature information of each detected defect, specifically including: defect type (such as porosity, inclusion, crack, etc.), confidence level of the type determination (0-100% probability value), three-dimensional spatial location of the defect in the steel plate, and predicted size (such as length, width, and area).

[0071] The preset rating criteria are specific clauses in the national standard GB / T 2970 or the enterprise's internal control standards regarding the correspondence between defect size, quantity, location and quality grade;

[0072] The high-precision verification module is connected in communication with the intelligent analysis module. It is used to cut out the corresponding sample steel plate according to the spatial location of the defect in the preliminary evaluation result, and to verify and measure it through high-precision detection methods to generate true defect data.

[0073] Furthermore, in the above technical solution, the high-precision verification module is as follows: Figure 4 The following are included:

[0074] The automatic sample cutting unit is used to cut sample steel plates containing specific defects from the production line according to the instructions of the intelligent analysis module.

[0075] The high-precision verification unit is an industrial CT scanning system used to scan the sample steel plate to obtain the defect type, size and defect level, and generate defect true value data.

[0076] It is important to understand that the automatic sample extraction unit is integrated into the rolling line, located downstream of the core detection unit. It mainly comprises a high-precision flying shear or laser cutter and a sample tray conveying system driven by a servo motor. Upon receiving instructions from the intelligent analysis module, the flying shear or laser cutter precisely positions the defect based on the steel plate position encoder information (X-coordinate) and the defect's lateral position (Y-coordinate) in the instructions. The PLC, based on the defect prediction size in the instructions, invokes a preset safety margin algorithm to automatically calculate the precise cutting path coordinates that include the defect and extend its boundary by a specific distance (e.g., 50mm). After calculation, the PLC sends the path coordinates to the actuator of the automatic sample extraction unit via a high-speed servo motion control bus (e.g., EtherCAT), driving the high-precision flying shear or laser cutter head to automatically extract the sample, thus ensuring that the defect is completely contained within the sample for subsequent CT scan verification.

[0077] Furthermore, in the above technical solution, refer to Figure 4 The triggering condition for the high-precision verification module is at least one of the following:

[0078] The system's continuous production time has reached the preset time cycle;

[0079] The system has cumulatively detected a number of steel plates that have reached a preset threshold.

[0080] The intelligent analysis module's initial assessment confidence level for a certain type of defect is lower than the preset confidence threshold.

[0081] It's important to know that the system's PLC has a built-in timer. When the continuous production time of the steel plate rolling line reaches a preset time cycle, such as every 8 hours of continuous production, the PLC sends a trigger command to the high-precision verification module. Each time the intelligent analysis module completes a steel plate inspection and preliminary evaluation, it sends a counting signal to the database. The counter in the database accumulates, and when the cumulative number of inspected steel plates reaches a preset value, such as 200 steel plates, the database management system, such as SQL Server, will automatically call a stored procedure to send a trigger command and the corresponding defect coordinate information to the high-precision verification module and the automatic sample extraction unit. When the rating unit of the intelligent analysis module outputs the preliminary evaluation results, it will judge the confidence level in real time. When the confidence level of a certain defect or type of defect, such as "delamination" or "inclusions," is lower than a preset threshold (e.g., confidence level < 85%), a high-priority instruction packet will be immediately generated and sent to the automatic sample extraction unit via industrial Ethernet. The instruction packet contains the precise spatial location code (X, Y, Z coordinates) of the defect.

[0082] The model optimization module is communicatively connected to the high-precision verification module and is used to optimize the artificial intelligence rating model based on the deviation between the defect truth data and the initial evaluation result.

[0083] Furthermore, in the above technical solution, refer to Figure 5 The model optimization module includes:

[0084] The deviation calculation unit is used to compare the initial evaluation results of the rating unit with the defect true value data generated by the high-precision verification unit, and calculate the type deviation, size deviation, rating deviation and comprehensive deviation degree.

[0085] The model optimization unit is used to perform backpropagation and parameter fine-tuning on the initial artificial intelligence rating model, using the comprehensive deviation as part of the loss function, to generate an optimized artificial intelligence rating model.

[0086] Furthermore, in the above technical solution, the deviation calculation unit is configured as follows:

[0087] The type deviation is calculated using the cross-entropy loss function;

[0088] The dimensional deviation is calculated using the mean square error.

[0089] The rating deviation is calculated using the grade difference;

[0090] The overall deviation is a function f(δ) type ,δ size ,δ level The calculation result of ), where δ type For type bias, δ size For dimensional deviation, δ level This indicates a rating deviation.

[0091] It should be noted that the deviation calculation unit is written in Python and uses the Pandas library to compare the initial assessment results with the true defect data. The type deviation calculation uses the cross-entropy loss function. Where p is the probability distribution of the true value type, q is the probability distribution of the initial evaluation type, and the calculation accuracy is ≤0.001; the mean square error is used for the dimensional deviation calculation. Where y is the true size, For the initial assessment dimensions, the calculation accuracy is ≤0.001mm; the rating deviation is calculated as the absolute difference between the initial assessment grade and the true grade. For example, if the initial assessment grade is level 2 and the true grade is level 1, the rating deviation L=1; the comprehensive deviation function is set as f(δ). type ,δ size ,δ level )=w1×δ type +w2×δ size +w3×δ levelWhere w1, w2, and w3 are weighting coefficients with a calculation precision of ≤0.001; the weighting coefficients w1, w2, and w3 are set by default to w1=0.5, w2=0.3, and w3=0.2, and are specifically determined according to the importance of each deviation item to the final quality rating.

[0092] The truth value type probability distribution p originates from the authoritative output of the high-precision verification module. Specifically, after the sample steel plate is subjected to high-precision verification and measurement by the industrial CT scanning system, the system generates defect truth value data, which has the final authority for the determination of defect type. In order to form the probability distribution p, the system converts this unique truth value type label into a discrete probability distribution, that is, the probability of the correct category is 1, and the probability of all other incorrect categories is 0.

[0093] The initial assessment type probability distribution q is directly derived from the original output of the artificial intelligence rating model in the intelligent analysis module. After processing the fully focused TFM image data transmitted from the online detection module, the model will directly output a probability distribution of various predefined defect types, which intuitively reflects the model's confidence in the type to which the current defect belongs.

[0094] The true value dimension y comes directly from the final measurement result of the high-precision verification module. After the sample steel plate is verified by the industrial CT scanning system, the defect true value data generated by the system will provide the precise physical size of the defect in three-dimensional space. This data is adopted by the system as the benchmark true value for evaluation.

[0095] The initial evaluation dimensions The result is output synchronously by the artificial intelligence rating model in the intelligent analysis module when reasoning on the fully focused TFM image data. It is a predicted estimate of the defect size made by the model based on its algorithm.

[0096] The model optimization unit uses the gradient descent optimization algorithm to backpropagate and fine-tune the parameters of the initial artificial intelligence rating model. This process uses the defect true data and its corresponding initial evaluation result deviation to adjust the weight parameters inside the model, minimize the comprehensive deviation, and finally generate an optimized artificial intelligence rating model with improved performance. The fine-tuning process is usually carried out on a small batch of accumulated validation samples to improve efficiency.

[0097] A dual quality control module, coupled with the intelligent analysis module, is used to deploy the optimized artificial intelligence rating model to the online detection module for subsequent steel plate detection, and to perform retrospective simulation screening and correction of historical detection data.

[0098] Furthermore, in the above technical solution, refer to Figure 6 The dual quality control module includes:

[0099] The forward-looking control unit is used to deploy the optimized artificial intelligence rating model to the intelligent analysis module to detect the steel plates that subsequently pass through the online detection module;

[0100] The retrospective control unit is used to call the optimized artificial intelligence rating model to recalculate and rate the stored historical detection data in order to identify and correct potential false detection events in the past and generate a correction report.

[0101] The historical detection data refers to the original historical ultrasound detection data.

[0102] In practice, the dual quality control module achieves continuous quality improvement through the collaborative operation of the forward-looking control unit and the retrospective control unit: whenever the model optimization module generates an optimized artificial intelligence rating model, the forward-looking control unit automatically deploys it to the intelligent analysis module on the online inspection line through a secure network channel, replacing the original model and making it immediately applicable to the real-time inspection of all steel plates passing through the production line, thereby achieving instant forward-looking iteration of inspection capabilities; at the same time, the retrospective control unit is triggered, which calls the optimized model to recalculate and rate all historical raw ultrasonic inspection data stored in the central database, especially the full matrix capture FMC data and steel plate position encoder information. By comparing the new rating results with the original records, it automatically identifies potential historical false detection events discovered due to model iteration, such as those that were originally judged as qualified but are now identified as defects, or situations where the original defect level assessment needs to be adjusted up or down; subsequently, the system automatically corrects the rating results and quality records of this part of the historical data and generates a detailed correction report, which includes the affected batch number, the unique identifier of the steel plate, defect location information, and the comparison results of the rating before and after the correction, and finally pushes it to the quality management department, thus forming a complete quality control closed loop covering future inspections and past data.

[0103] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fully automated online detection system for internal defects in steel plates based on ultrasonic phased array, characterized in that, include: The online inspection module is used to perform non-destructive testing on moving steel plates on the steel plate rolling line and generate raw ultrasonic test data; The online detection module includes: The mechanical conveying and centering unit is used to convey steel plates and ensure that they are centered as they pass through the inspection area; The coupling unit is used to provide a stable coupling medium between the ultrasonic testing probe and the steel plate surface; The core detection unit is used to collect and generate raw ultrasound detection data; The intelligent analysis module is communicatively connected to the online detection module and is used to run an artificial intelligence rating model to process and preliminarily evaluate the original ultrasonic detection data, and output a preliminary evaluation result including the spatial location of the defects. The intelligent analysis module includes: The defect intelligent identification unit is used to receive the raw ultrasonic test data from the online detection module, process the raw ultrasonic test data to obtain complete ultrasonic test data, input the complete ultrasonic test data into the initial artificial intelligence rating model, and output the physical characteristic information of the defect, including defect type, confidence level, spatial location and predicted size. The rating unit is used to output the rating result of the defect based on the physical characteristics of the defect and according to the preset rating criteria, which together constitute the initial rating result. The high-precision verification module is connected in communication with the intelligent analysis module. It is used to cut out the corresponding sample steel plate according to the spatial location of the defect in the preliminary evaluation result, and to verify and measure it through high-precision detection methods to generate true defect data. The high-precision verification module includes: The automatic sample cutting unit is used to cut sample steel plates containing specific defects from the production line according to the instructions of the intelligent analysis module. The high-precision verification unit is an industrial CT scanning system used to scan the sample steel plate to obtain the defect type, size and defect level, and generate defect true value data; The model optimization module is communicatively connected to the high-precision verification module and is used to optimize the artificial intelligence rating model based on the deviation between the defect truth data and the initial evaluation result. The model optimization module includes: The deviation calculation unit is used to compare the initial evaluation results of the rating unit with the defect true value data generated by the high-precision verification unit, and calculate the type deviation, size deviation, rating deviation and comprehensive deviation degree. The model optimization unit is used to perform backpropagation and parameter fine-tuning on the initial artificial intelligence rating model, using the comprehensive deviation as part of the loss function, to generate an optimized artificial intelligence rating model. A dual quality control module, coupled with the intelligent analysis module, is used to deploy the optimized artificial intelligence rating model to the online detection module for subsequent steel plate detection, and to perform retrospective simulation screening and correction of historical detection data. The dual quality control module includes: The forward-looking control unit is used to deploy the optimized artificial intelligence rating model to the intelligent analysis module to detect the steel plates that subsequently pass through the online detection module; The retrospective control unit is used to call the optimized artificial intelligence rating model to recalculate and rate the stored historical detection data in order to identify and correct potential false detection events in the past and generate a correction report.

2. The fully automated online detection system for internal defects of steel plates based on ultrasonic phased array as described in claim 1, characterized in that, The raw ultrasonic testing data includes full matrix capture (FMC) data and steel plate position encoder information.

3. The fully automated online detection system for internal defects of steel plates based on ultrasonic phased array according to claim 1, characterized in that, The triggering condition for the high-precision verification module is at least one of the following: The system's continuous production time has reached the preset time cycle; The system has cumulatively detected a number of steel plates that have reached a preset threshold. The intelligent analysis module's initial assessment confidence level for a certain type of defect is lower than the preset confidence threshold.

4. The fully automated online detection system for internal defects in steel plates based on ultrasonic phased array as described in claim 1, characterized in that, The deviation calculation unit is configured as follows: The type deviation is calculated using the cross-entropy loss function; The dimensional deviation is calculated using the mean square error. The rating deviation is calculated using the grade difference; The overall deviation is a function f(δ) type ,δ size ,δ level The calculation result of ), where δ type For type bias, δ size For dimensional deviation, δ level This indicates a rating deviation.

Images