Casting size automatic detection system

Abstract

The invention relates to the technical field of casting size detection, in particular to an automatic casting size detection system. According to the invention, through cooperation of the multi-stage image acquisition module and the acquisition control analysis module, the image acquisition resolution is dynamically adjusted according to the comprehensive abnormal value; when the abnormal condition is small, low-resolution acquisition is adopted, the data processing amount is reduced, and the detection speed is increased; when abnormal conditions are aggravated, the resolution is improved, and the detection precision is ensured, so that the detection precision and the production line efficiency are balanced, and the productivity loss caused by high-resolution acquisition is avoided; production process data and equipment state data are fused through the acquisition control analysis module, a comprehensive abnormal value is calculated through multi-source data weighted fusion, a corresponding acquisition control instruction is generated, full-link abnormities of production can be monitored in real time, potential problems can be pre-warned in advance, the detection accuracy is improved, the system operation stability is optimized, and the product quality is improved. And the false detection rate caused by equipment or process abnormity is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of casting size detection, in particular to an automatic casting size detection system. Background Art

[0002] Automotive castings are the key foundation for core components such as engine blocks and transmission housings. Their quality directly affects the power, reliability, and safety of the entire vehicle. Therefore, in the production of automotive castings, the accuracy of their dimensions plays a key role, and the detection of casting dimensions is particularly important. Existing casting dimension monitoring requires image acquisition through high-precision sensors for dimension detection. However, in order to accurately detect casting dimensions, high-resolution images need to be collected. However, the processing and transmission of such images is time-consuming, which can easily cause the detection speed to lag behind the production line rhythm, resulting in reduced production capacity. Summary of the Invention

[0003] The present invention provides a casting size automatic detection system for solving the above technical problems.

[0004] A first aspect of the present invention provides a casting size automatic detection system, comprising a multi-stage image acquisition module, a size calculation module, an acquisition control and analysis module, and a main control communication module.

[0005] The multi-level image acquisition module includes a multi-level image acquisition layer and an acquisition decision layer. The acquisition decision layer receives acquisition control instructions and generates acquisition decisions based on the acquisition control instructions, and obtains the casting positioning signal. The multi-level image acquisition layer identifies the casting positioning signal. When the casting positioning signal corresponds to ready, image acquisition is executed on the corresponding casting based on the acquisition decision. The casting image data is obtained through the acquisition execution and sent to the size calculation module; the casting image data includes visible light image data and three-dimensional point cloud data.

[0006] As a further improvement of the present invention, a collection decision is generated according to the collection control instruction, specifically: the collection control instruction is identified, and the collection control instruction includes a low-precision requirement instruction, a medium-precision requirement instruction, a high-precision requirement instruction, and a collection warning instruction; when the collection control instruction corresponds to the collection warning instruction, a shutdown warning message and a corresponding collection stop instruction are generated; when the collection control instruction corresponds to the low, medium, and high-precision requirement instructions, the collection resolutions corresponding to the multi-level image collection layers are divided into low, medium, and high-resolution intervals corresponding to the low, medium, and high-precision requirement instructions, the comprehensive anomaly value corresponding to the collection control instruction and the corresponding comprehensive anomaly threshold interval are obtained, the difference calculation is performed according to the maximum value and the minimum value of the comprehensive anomaly threshold interval to obtain the interval fluctuation range value, the difference calculation is performed between the corresponding comprehensive anomaly value and the corresponding interval fluctuation range value to obtain the interval occupancy ratio, the interval occupancy ratio is matched with the corresponding resolution interval to obtain the corresponding required resolution, and the required resolution is used as the corresponding collection decision.

[0007] The size calculation module is used to input the received casting image data into the size calculation layer, and the size calculation layer performs size calculation and analysis on the casting image data to obtain the casting size state.

[0008] As a further improvement of the present invention, the size calculation and analysis of the casting image data are as follows:

[0009] Visible light image data and three-dimensional point cloud data are obtained based on the casting image data;

[0010] Pixel coordinate information is obtained based on the visible light image data, and the image pixel coordinates are obtained according to the pixel coordinate information and denoted as (u yc , , v i ), and the camera calibration and coordinate transformation model is used to convert each image pixel coordinate into world coordinates (X i , Y i , Z i ); where s represents a preset scale factor, represents the camera internal parameter matrix, f x , f y are focal lengths, (u0, v0) are the principal point coordinates; [R|t] represents the external parameter matrix, R is the rotation matrix, and t is the translation vector; the image pixel coordinates and world coordinates corresponding to the contour points of the casting are obtained, and the least squares circle fitting of the contour fitting deviation algorithm is used for the world coordinates corresponding to the contour points to obtain the fitting center coordinates (a, b), where r represents the radius; the circular size deviation C [[ID=3​​​​​​are all preset theoretical center coordinates; through the measured length calculation formula the measured length L is calculated 实测 ; the linear dimension deviation is obtained by calculating the difference between the measured length and the preset theoretical length.

[0011] Based on the three-dimensional point cloud data, the measured point cloud data is obtained, and the transformation matrix T0 = [R0|t0] is obtained through the SAC-IA algorithm; the iterative optimization formula is obtained by using the iterative closest point algorithm where p i represents the measured point cloud data; represents the nearest neighbor point of p in the theoretical point cloud Q i ; the root mean square error value is calculated through the root mean square error formula

[0012] The circular dimension deviation, linear dimension deviation and root mean square error value are normalized and their numerical values are taken, and the comprehensive deviation index Ipcz is calculated by using the exponential weighted cross model ; where α1, α2, α3, α4, α5, α6 are all preset weight coefficients and adjustment parameters, α1, α3, α4, α6 are all greater than zero, and α2, α5 are all greater than one; the pre-designed comprehensive deviation threshold is obtained. When the comprehensive deviation index exceeds the comprehensive deviation threshold, the corresponding casting size state is generated as an abnormal casting; otherwise, the corresponding casting size state is generated as a normal casting.

[0013] The acquisition control analysis module is used to obtain the casting size state of each casting corresponding to the preset analysis period, and perform acquisition control analysis on the casting size state corresponding to the analysis period to obtain an acquisition control instruction, and send the acquisition control instruction to the multi-level image acquisition module.

[0014] As a further improvement of the present invention, the acquisition control analysis is performed on the casting size state corresponding to the analysis period, and the specific analysis content is as follows:

[0015] Identify the casting size state within the analysis period, count the number of abnormal castings corresponding to the abnormal castings, calculate the ratio of the number of abnormal castings to the total number of castings in the time-sharing period to obtain the abnormal casting rate. When the abnormal casting rate exceeds the preset abnormal casting threshold, calculate the difference between the abnormal casting rate and the abnormal casting threshold to obtain the abnormal casting index; when the abnormal casting index is detected, obtain the corresponding comprehensive deviation index, calculate the difference between the comprehensive deviation index and the comprehensive deviation threshold to obtain the deviation overflow value, and calculate the product of the deviation overflow value and the abnormal casting index to obtain the detection abnormal value.

[0016] ​Based on the database, we can obtain production process data and equipment status data; obtain pouring temperature data and mold wear data according to the production process data; obtain real-time temperature according to the pouring temperature data, and use the thermodynamic hysteresis model to calculate the actual temperature. Calculate the pouring temperature abnormality K T ; Among them, T t 、T pred are real-time temperature and temperature predicted by thermodynamic model respectively; T bz It is represented by the preset temperature standard deviation; k and η represent the cooling coefficient and lag time respectively;

[0017] The mold state mt is obtained based on the mold wear data. The mold state corresponds to no wear, slight wear and severe wear. The wear amount observation sequence before the preset time is obtained and the state transition of the hidden Markov model is used. Calculate the wear abnormality K m ;in, π i They are respectively represented as the wear amount observation sequence before time t and the preset initial probability of the state; Expressed as the posterior probability.

[0018] Obtain robot arm positioning data and sensor noise data based on device status data; obtain real-time positioning trajectory and reference positioning trajectory based on robot arm positioning data; substitute the real-time positioning trajectory and reference positioning trajectory into the dynamic time warping distance model Calculate the positioning anomaly degree K J ; Among them, d t 、 They represent the real-time positioning trajectory and the reference positioning trajectory respectively; DTW is the dynamic time warping distance, DTW min and DTW max They are respectively expressed as the corresponding historical minimum distance and historical maximum distance; when the positioning anomaly degree is greater than the preset threshold, the positioning anomaly degree is calculated by subtracting the preset threshold to obtain the positioning anomaly index;

[0019] Based on the sensor noise data of each sensor, wavelet packet decomposition is performed to obtain multiple wavelet packet decomposition layers, and the sub-band energy corresponding to each wavelet packet decomposition layer is obtained. The wavelet packet energy entropy model is used The noise anomaly degree N is calculated; Nc is the number of wavelet packet decomposition layers, r jThe noise anomaly degree is expressed as the j-th subband energy; the noise anomaly degree is divided into multiple noise anomaly degree intervals, and the intervals are arranged in ascending order based on the numerical value of the maximum anomaly degree of each noise anomaly degree interval to generate a corresponding interval serial number, and an increasing noise influence coefficient is given based on the interval serial number; the current corresponding noise anomaly degree is matched with each noise anomaly degree interval to obtain the corresponding noise influence coefficient, and the noise anomaly degree is multiplied by the noise influence coefficient to obtain the sensor anomaly value.

[0020] The monitoring abnormal values, pouring temperature abnormality, wear abnormality, positioning abnormality index and sensor abnormal values ​​are normalized and their values ​​are taken, and all are recorded as data abnormal values; various types of data abnormal values ​​are fused using the multi-source data weighted fusion formula Calculate the comprehensive outlier value K zong ; Among them, Ki represents the data outlier of the i-th category, θi represents the data weight factor corresponding to the data outlier of the i-th category, τ i , t i They represent the preset time attenuation coefficient and abnormal duration corresponding to the abnormal value of the i-th type of data respectively; obtain the pre-designed comprehensive abnormality threshold, which includes the maximum comprehensive abnormality threshold and the minimum comprehensive abnormality threshold, respectively denoted as Kmax and Kmin; based on the comprehensive abnormality threshold, the comprehensive abnormality threshold intervals [0, Kmin], (Kmin, Kmax], (Kmax, Kz] are obtained; where Kz represents the total amount of abnormal allowable value; when K zong ∈[0, Kmin], the corresponding acquisition control instruction is generated as a low-precision demand instruction; when K zong ∈(Kmin, Kmax], the corresponding acquisition control instruction is generated as the medium precision requirement instruction; when K zong ∈(Kmax, Kz], the corresponding acquisition control instruction is generated as a high-precision demand instruction; when K zong When >Kz, the corresponding acquisition control instruction is generated as the acquisition warning instruction.

[0021] The main control communication module is used to establish a data transmission link for each module and store the transmitted casting image data, casting size status and acquisition control instructions in the database.

[0022] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0023] 1. The present invention dynamically adjusts the image acquisition resolution based on the comprehensive abnormality value through the collaboration of a multi-level image acquisition module and an acquisition control and analysis module. When the abnormality is minor, low-resolution acquisition is adopted to reduce the data processing volume and improve the detection speed. When the abnormality intensifies, the resolution is increased to ensure the detection accuracy, thereby balancing the detection accuracy and production line efficiency and avoiding the loss of production capacity caused by high-resolution acquisition.

[0024] 2. The present invention integrates production process data and equipment status data through the acquisition control analysis module, calculates comprehensive abnormal values ​​through weighted fusion of multi-source data, and generates corresponding acquisition control instructions. It can monitor abnormalities in all production links in real time, warn of potential problems in advance, improve detection accuracy, optimize system operation stability, and reduce false detection rates caused by equipment or process abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present application.

[0026] Figure 1 This is a principle block diagram of an automatic casting size detection system of the present invention. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0028] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 In an embodiment of the present invention, an embodiment of a casting size automatic detection system includes:

[0029] The multi-level image acquisition module includes a multi-level image acquisition layer and an acquisition decision layer. The acquisition decision layer receives acquisition control instructions and generates acquisition decisions based on the acquisition control instructions, and obtains the casting positioning signal. The multi-level image acquisition layer identifies the casting positioning signal. When the casting positioning signal corresponds to ready, image acquisition is executed on the corresponding casting based on the acquisition decision. The casting image data is obtained through acquisition execution and sent to the size calculation module; the casting image data includes visible light image data and three-dimensional point cloud data; the multi-level image acquisition layer is composed of multiple layers of image acquisition devices with different resolutions or devices with adjustable resolution, and can execute image acquisition of different precisions according to the acquisition decision.

[0030] Generate an acquisition decision according to the acquisition control instruction, which is specifically as follows: identify the acquisition control instruction, which includes low-precision requirement instructions, medium-precision requirement instructions, high-precision requirement instructions and acquisition warning instructions; when the acquisition control instruction corresponds to the acquisition warning instruction, generate shutdown warning information and the corresponding acquisition stop instruction; when the acquisition control instruction corresponds to the low, medium and high-precision requirement instructions, divide the acquisition resolution corresponding to the multi-level image acquisition layer into low, medium and high-resolution intervals corresponding to the low, medium and high-precision requirement instructions, obtain the comprehensive abnormality value and the corresponding comprehensive abnormality threshold interval corresponding to the acquisition control instruction, calculate the difference between the maximum and minimum values ​​of the comprehensive abnormality threshold interval to obtain the interval fluctuation range value, calculate the difference between the corresponding comprehensive abnormality value and the corresponding interval fluctuation range value to obtain the interval proportion value, match the interval proportion value with the corresponding resolution interval to obtain the corresponding required resolution, and use the required resolution as the corresponding acquisition decision; generate differentiated acquisition decisions according to different acquisition control instructions, shut down when encountering a warning instruction, match the resolution according to the degree of abnormality when normal, and improve the system response flexibility.

[0031] For example, when the abnormal casting rate of a certain automobile casting production line exceeds the threshold within the preset analysis period, the acquisition control analysis module calculates the abnormal casting index and the deviation overflow value to obtain the detection abnormality value. At the same time, combined with the pouring temperature abnormality, the mold slight wear abnormality, the robot arm positioning abnormality index and the sensor noise abnormality, the comprehensive abnormality value Kzong>Kz is obtained through weighted fusion of multi-source data. At this time, an acquisition warning instruction is generated. After receiving it, the multi-level image acquisition module stops acquiring and issues a shutdown warning to avoid the production of batches of unqualified castings. At the same time, the abnormal data is traced through the database to assist in process optimization.

[0032] The size calculation module inputs the received casting image data into the size calculation layer, and the size calculation layer performs size calculation and analysis on the casting image data to obtain the casting size status.

[0033] The size calculation and analysis of the casting image data is carried out, and the specific analysis contents are as follows:

[0034] Obtain visible light image data and three-dimensional point cloud data based on casting image data;

[0035] Based on the visible light image data, the pixel coordinate information is obtained, and the image pixel coordinates are obtained according to the pixel coordinate information and recorded as (u i , v i ), using camera calibration and coordinate transformation model Convert each image pixel coordinate to world coordinate (X i , Y i , Z i ), where s represents the preset scale factor, Expressed as the camera intrinsic parameter matrix, fx , f y is the focal length, (u0, v0) are the principal point coordinates; [R|t] represents the external parameter matrix, R is the rotation matrix, and t is the translation vector; obtain the image pixel coordinates and world coordinates corresponding to the contour points of the casting, and use the least squares circle fitting of the contour fitting deviation algorithm for the world coordinates corresponding to the contour points to obtain the fitted center coordinates (a, b), where r represents the radius; calculate the circular dimension deviation C through the circular dimension deviation calculation formula yc ; where a 理论 and b 理论 are both preset theoretical center coordinates; calculate the measured length L through the measured length calculation formula 实测 ; Calculate the difference between the measured length and the preset theoretical length to obtain the linear dimension deviation.

[0036] Based on the three-dimensional point cloud data, obtain the measured point cloud data, and obtain the transformation matrix T0 = [R0|t0] through the SAC-IA algorithm; use the iterative closest point algorithm to obtain the iterative optimization formula where p i represents the measured point cloud data; represents the nearest neighbor point of p i in the theoretical point cloud Q; calculate the root mean square error value through the root mean square error formula

[0037] Normalize the circular dimension deviation, linear dimension deviation, and root mean square error value and take their numerical values, and use the exponential weighted cross model to calculate the comprehensive deviation index Ipcz; where α1, α2, α3, α4, α5, α6 are all preset weight coefficients and adjustment parameters, α1, α3, α4, α6 are all greater than zero, and α2, α5 are all greater than one; obtain the preset comprehensive deviation threshold, and when the comprehensive deviation index exceeds the comprehensive deviation threshold, generate the corresponding casting size status as an abnormal casting; otherwise, generate the corresponding casting size status as a normal casting; combine the visible light image and the three-dimensional point cloud data, calculate the size deviation through multiple algorithms, and evaluate the casting status from multiple dimensions to improve the reliability of the detection results.

[0038] The acquisition control analysis module obtains the casting size status of each casting corresponding to the preset analysis period, and performs acquisition control analysis on the casting size status corresponding to the analysis period to obtain the acquisition control instruction, and sends the acquisition control instruction to the multi-level image acquisition module.

[0039] Perform acquisition control analysis on the casting size status corresponding to the analysis period, and its specific analysis content is as follows:

[0040] ​​​The casting size status within the analysis period is identified, and the number of abnormal castings corresponding to the abnormal castings is obtained by statistics. The abnormal casting rate is calculated by ratio of the abnormal casting number to the total number of castings in the time-sharing period. When the abnormal casting rate exceeds the pre-set abnormal casting threshold, the abnormal casting rate and the abnormal casting threshold are subtracted to obtain the abnormal casting index; when the abnormal casting index is monitored, the corresponding comprehensive deviation index is obtained, the comprehensive deviation index and the comprehensive deviation threshold are subtracted to obtain the deviation overflow value, and the deviation overflow value is multiplied by the abnormal casting index to obtain the detection abnormality value.

[0041] Based on the database, we can obtain production process data and equipment status data; obtain pouring temperature data and mold wear data according to the production process data; obtain real-time temperature according to the pouring temperature data, and use the thermodynamic hysteresis model to calculate the actual temperature. Calculate the pouring temperature abnormality K T ; Among them, T t 、T pred are real-time temperature and temperature predicted by thermodynamic model respectively; T bz It is expressed as the preset temperature standard deviation; k and η represent the cooling coefficient and lag time respectively; the thermodynamic lag model analyzes the pouring temperature anomaly model, taking into account the lag relationship between the real-time temperature and the predicted temperature

[0042] The mold state mt is obtained based on the mold wear data. The mold state corresponds to no wear, slight wear and severe wear. The wear amount observation sequence before the preset time is obtained and the state transition of the hidden Markov model is used. Calculate the wear abnormality K m ;in, π i They are respectively represented as the wear amount observation sequence before time t and the preset initial probability of the state; It is expressed as a posterior probability; the hidden Markov model analyzes the mold wear state transition through the wear amount observation sequence and calculates the wear abnormality.

[0043] Obtain robot arm positioning data and sensor noise data based on device status data; obtain real-time positioning trajectory and reference positioning trajectory based on robot arm positioning data; substitute the real-time positioning trajectory and reference positioning trajectory into the dynamic time warping distance model Calculate the positioning anomaly degree K J ; Among them, d t 、 They represent the real-time positioning trajectory and the reference positioning trajectory respectively; DTW is the dynamic time warping distance, DTW min and DTW maxThey are respectively expressed as the corresponding historical minimum distance and historical maximum distance; when the positioning anomaly degree is greater than the preset threshold, the positioning anomaly degree is calculated by subtracting the preset threshold to obtain the positioning anomaly index;

[0044] Based on the sensor noise data of each sensor, wavelet packet decomposition is performed to obtain multiple wavelet packet decomposition layers, and the sub-band energy corresponding to each wavelet packet decomposition layer is obtained. The wavelet packet energy entropy model is used The noise anomaly degree N is calculated; Nc is the number of wavelet packet decomposition layers, r j It is expressed as the j-th subband energy; wavelet packet decomposition is an extension of wavelet transform, which performs a more refined multi-layer decomposition of the signal in the frequency domain, and can process both low-frequency and high-frequency components; the noise anomaly degree is divided into multiple noise anomaly degree intervals, and the numerical value of the maximum anomaly degree of each noise anomaly degree interval is sorted in ascending order to generate the corresponding interval serial number, and an increasing noise influence coefficient is given based on the interval serial number; the current corresponding noise anomaly degree is matched with each noise anomaly degree interval to obtain the corresponding noise influence coefficient, and the noise anomaly degree is multiplied by the noise influence coefficient to obtain the sensor anomaly value.

[0045] The monitoring abnormal values, pouring temperature abnormality, wear abnormality, positioning abnormality index and sensor abnormal values ​​are normalized and their values ​​are taken, and all are recorded as data abnormal values; various types of data abnormal values ​​are fused using the multi-source data weighted fusion formula Calculate the comprehensive outlier value K zong ; Among them, Ki represents the data outlier of the i-th category, θi represents the data weight factor corresponding to the data outlier of the i-th category, τ i , t i They represent the preset time attenuation coefficient and abnormal duration corresponding to the abnormal value of the i-th type of data respectively; obtain the pre-designed comprehensive abnormality threshold, which includes the maximum comprehensive abnormality threshold and the minimum comprehensive abnormality threshold, respectively denoted as Kmax and Kmin; based on the comprehensive abnormality threshold, the comprehensive abnormality threshold intervals [0, Kmin], (Kmin, Kmax], (Kmax, Kz] are obtained; where Kz represents the total amount of abnormal allowable value; when K zong ∈[0, Kmin], the corresponding acquisition control instruction is generated as a low-precision demand instruction; when K zong ∈(Kmin, Kmax], the corresponding acquisition control instruction is generated as the medium precision requirement instruction; when K zong ∈(Kmax, Kz], the corresponding acquisition control instruction is generated as a high-precision demand instruction; when K zong When >Kz, the corresponding acquisition control instruction is generated as the acquisition warning instruction.

[0046] The main control communication module establishes a data transmission link for each module and stores the transmitted casting image data, casting size status and acquisition control instructions in the database; it builds a stable data transmission and storage system to provide support for the coordinated work of each module in the system, and accumulates data resources for subsequent production optimization.

[0047] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A casting size automatic detection system, characterized in that: include: The multi-level image acquisition module includes a multi-level image acquisition layer and an acquisition decision layer. The acquisition decision layer receives acquisition control instructions and generates acquisition decisions based on the acquisition control instructions, and obtains casting positioning signals. The multi-level image acquisition layer identifies the casting positioning signals. When the casting positioning signals correspond to ready, image acquisition is executed for the corresponding casting based on the acquisition decision. The casting image data is obtained through the acquisition execution and sent to the size calculation module. The size calculation module is used to input the received casting image data into the size calculation layer, and the size calculation layer performs size calculation and analysis on the casting image data to obtain the casting size status; The acquisition control analysis module is used to obtain the casting size status of each casting corresponding to the preset analysis period, and perform acquisition control analysis on the casting size status corresponding to the analysis period to obtain acquisition control instructions, and send the acquisition control instructions to the multi-level image acquisition module.

2. The automatic casting size detection system according to claim 1, characterized in that: Generate an acquisition decision according to the acquisition control instruction, which is specifically: The acquisition control instructions are identified, and the acquisition control instructions include low-precision requirement instructions, medium-precision requirement instructions, high-precision requirement instructions and acquisition warning instructions; when the acquisition control instruction corresponds to the acquisition warning instruction, a shutdown warning information and a corresponding acquisition stop instruction are generated; when the acquisition control instruction corresponds to the low, medium and high-precision requirement instructions, the acquisition resolution corresponding to the multi-level image acquisition layer is divided into low, medium and high-resolution intervals corresponding to the low, medium and high-precision requirement instructions, and the comprehensive abnormality value and the corresponding comprehensive abnormality threshold interval corresponding to the acquisition control instruction are obtained. The interval fluctuation range value is obtained by performing a difference calculation based on the maximum and minimum values ​​of the comprehensive abnormality threshold interval, and the interval proportion value is obtained by performing a difference calculation between the corresponding comprehensive abnormality value and the corresponding interval fluctuation range value. The interval proportion value is matched with the corresponding resolution interval to obtain the corresponding required resolution, and the required resolution is used as the corresponding acquisition decision.

3. The automatic casting size detection system according to claim 1, characterized in that: The casting image data is subjected to dimensional calculation and analysis, and the specific analysis contents are as follows: Obtain visible light image data and three-dimensional point cloud data based on casting image data; Obtain pixel coordinate information based on visible light image data, obtain image pixel coordinates based on the pixel coordinate information, and convert each image pixel coordinate into a world coordinate using a camera calibration and coordinate conversion model; obtain the image pixel coordinates and world coordinates corresponding to the contour point corresponding to the casting, and use the least squares circle fitting method of the contour fitting deviation algorithm to obtain the fitting circle center coordinates for the world coordinates corresponding to the contour point; calculate the circle size deviation using the circle size deviation calculation formula; then calculate the measured length using the measured length calculation formula; and calculate the difference between the measured length and the preset theoretical length to obtain the linear size deviation; Based on the three-dimensional point cloud data, the measured point cloud data is obtained, the transformation matrix is ​​obtained through the SAC-IA algorithm, and the iterative optimization formula is obtained using the iterative closest point algorithm; The root mean square error value is calculated using the root mean square error formula based on the information of the transformation matrix and the iterative optimization formula; The circular dimension deviation, linear dimension deviation and root mean square error values ​​are normalized and their values ​​are taken, and the comprehensive deviation index is calculated using the exponential weighted cross model; a pre-designed comprehensive deviation threshold is obtained. When the comprehensive deviation index exceeds the comprehensive deviation threshold, the corresponding casting dimension status is generated as an abnormal casting; otherwise, the corresponding casting dimension status is generated as a normal casting.

4. The automatic casting size detection system according to claim 1, characterized in that: The casting size status corresponding to the analysis period is collected, controlled and analyzed, and the specific analysis contents are as follows: The casting size status within the analysis period is identified, and the number of abnormal castings corresponding to the abnormal castings is obtained by counting. The abnormal casting rate is calculated by the ratio of the abnormal casting number to the total number of castings in the time-sharing period. When the abnormal casting rate exceeds the preset abnormal casting threshold, the abnormal casting rate and the abnormal casting threshold are subtracted to obtain the abnormal casting index. When the abnormal casting index is monitored, the corresponding comprehensive deviation index is obtained, and the comprehensive deviation index and the comprehensive deviation threshold are subtracted to obtain the deviation overflow value. The deviation overflow value is multiplied by the abnormal casting index to obtain the detection abnormality value. Production process data and equipment status data are obtained based on the database; pouring temperature data and mold wear data are obtained based on the production process data, and the pouring temperature data and mold wear data are analyzed to obtain pouring temperature abnormality and wear abnormality; robot positioning data and sensor noise data are obtained based on the equipment status data, and the robot positioning data and sensor noise data are analyzed to obtain positioning abnormality index and sensor abnormality value; comprehensive analysis is performed on monitoring abnormality values, pouring temperature abnormality, wear abnormality, positioning abnormality index and sensor abnormality value to obtain acquisition control instructions.

5. The automatic casting size detection system according to claim 4, characterized in that: The pouring temperature data and mold wear data are analyzed, and the specific analysis is as follows: The real-time temperature is obtained based on the pouring temperature data, and the pouring temperature anomaly is calculated using the thermodynamic hysteresis model. The mold state is obtained based on the mold wear data, and the mold state corresponds to no wear, slight wear, and severe wear. The wear amount observation sequence before the preset time is obtained, and the hidden Markov model state transition is used to analyze the mold state to obtain the wear anomaly.

6. The automatic casting size detection system according to claim 4, characterized in that: The robot arm positioning data and sensor noise data are analyzed, and the specific analysis is as follows: The real-time positioning trajectory and the reference positioning trajectory are obtained based on the positioning data of the robotic arm; the real-time positioning trajectory and the reference positioning trajectory are substituted into the dynamic time warping distance model to calculate the positioning anomaly degree; when the positioning anomaly degree is greater than a preset threshold, the positioning anomaly degree is subtracted from the preset threshold to obtain a positioning anomaly index; wavelet packet decomposition is performed based on the sensor noise data of each sensor to obtain multiple wavelet packet decomposition layers, the subband energy corresponding to each wavelet packet decomposition layer is obtained, and the noise anomaly degree is calculated and analyzed using the wavelet packet energy entropy model; The noise anomaly degree is divided into multiple noise anomaly degree intervals, and the values ​​of the maximum anomaly degrees of each noise anomaly degree interval are sorted in ascending order to generate corresponding interval serial numbers. An increasing noise impact coefficient is given based on the interval serial numbers. The current corresponding noise anomaly degree is matched with each noise anomaly degree interval to obtain the corresponding noise impact coefficient. The noise anomaly degree and the noise impact coefficient are multiplied to obtain the sensor anomaly value.

7. The automatic casting size detection system according to claim 1, characterized in that: A comprehensive analysis is performed on the monitoring abnormal values, pouring temperature abnormality, wear abnormality, positioning abnormality index and sensor abnormal values, specifically: The monitoring abnormal values, pouring temperature abnormality, wear abnormality, positioning abnormality index and sensor abnormal values ​​are all normalized and their values ​​are taken, and are all recorded as data abnormal values; various types of data abnormal values ​​are calculated using the multi-source data weighted fusion formula to obtain the comprehensive abnormal value; the pre-designed comprehensive abnormality threshold is obtained, and the comprehensive abnormality threshold includes the maximum comprehensive abnormality threshold and the minimum comprehensive abnormality threshold, which are recorded as Kmax and Kmin respectively; based on the comprehensive abnormality threshold, the comprehensive abnormality threshold interval [0, Kmin], (Kmin, Kmax], (Kmax, Kz] is obtained; Kz represents the total amount of abnormal allowable value; when K zong ∈[0, Kmin], the corresponding acquisition control instruction is generated as a low-precision demand instruction; when K zong ∈(Kmin, Kmax], the corresponding acquisition control instruction is generated as the medium precision requirement instruction; when K zong ∈(Kmax, Kz], the corresponding acquisition control instruction is generated as a high-precision demand instruction; when K zong When >Kz, the corresponding acquisition control instruction is generated as the acquisition warning instruction.

8. The automatic casting size detection system according to claim 1, characterized in that: It also includes a main control communication module, which is used to establish a data transmission link for each module and store the transmitted casting image data, casting size status and acquisition control instructions in the database; the casting image data includes visible light image data and three-dimensional point cloud data.

Images