Water pump shell image processing method and system for quality detection

Abstract

The invention relates to a water pump shell image processing method and system for quality detection, and the method comprises the steps: obtaining a production control parameter deviation vector matrix through comparing a production expectation control parameter and a production monitoring control parameter of a water pump shell; searching a high-frequency water pump shell defect and a low-frequency water pump shell defect which meet the production control parameter deviation vector matrix and of which the trigger frequencies are greater than or equal to a trigger frequency threshold value from a water pump shell historical sample; establishing a high-frequency feature extractor and a low-frequency feature extractor based on a predefined defect feature extractor calibration table; performing feature extraction on the water pump shell image meeting the condition to obtain an image feature extraction result; the technical effects of saving inspection resources and improving the inspection efficiency are achieved.

Description

Technical Field

[0001] The present invention relates to the field of detection technologies, and particularly to an image processing method and system for a water pump housing used for quality detection. Background Art

[0002] In the traditional production process of water pump housings, the processing schemes for defect monitoring and image feature extraction usually adopt a fixed process, using the same detection process for all housing types, and unable to dynamically adjust the detection process according to the changes in each link and factor during production. As a result, a large amount of detection resources are wasted on detection items with low importance, and the detection parameters that really need attention are easily overlooked, leading to waste of inspection resources and low efficiency. Summary of the Invention

[0003] In view of the technical problems of waste of inspection resources and low inspection efficiency in the prior art, the present invention provides an image processing method for a water pump housing used for quality detection to solve these problems.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: In a first aspect, the present invention provides an image processing method for a water pump housing used for quality detection, including: Comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix; retrieving high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and have a triggering frequency greater than or equal to a triggering frequency threshold from the historical samples of the water pump housing; processing the high-frequency water pump housing defects and the low-frequency water pump housing defects based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor; invoking the high-frequency feature extractor to extract features from all water pump housing images that meet the production control parameter deviation vector matrix, and invoking the low-frequency feature extractor to extract features from a set proportion of water pump housing images that meet the production control parameter deviation vector matrix to obtain an image feature extraction result.

[0005] Optionally, comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix includes: Normalizing the production expected control parameters and the production monitoring control parameters respectively to obtain production expected control eigenvalues and production monitoring control eigenvalues; Calculating the absolute deviation values of the same time domain and the same attribute for the production expected control eigenvalues and the production monitoring control eigenvalues to obtain the first attribute deviation time series information to the Nth attribute deviation time series information, where the time series covers the entire production time zone and the matrix elements are configured as empty when the attribute is not set at the corresponding moment.

[0006] Optionally, retrieve high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and have a trigger frequency greater than or equal to the trigger frequency threshold from the water pump housing historical samples. The trigger frequency threshold includes a first trigger frequency threshold and a second trigger frequency threshold, and the first trigger frequency threshold is greater than the second trigger frequency threshold. This includes: Extract the sample production control parameter deviation vector matrix from the water pump housing historical samples; Calculate the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix. If it is greater than or equal to the matrix similarity threshold, add it to the selected water pump housing historical samples. Otherwise, update the water pump housing historical samples; When the number of the selected water pump housing historical samples is greater than or equal to the set number, count the trigger frequencies of the first defect attribute until the Qth defect attribute, where the set number ≥ 10000 and all historical samples belong to the set time window; Based on the trigger frequencies of the first defect attribute until the Qth defect attribute, extract the defects with attribute trigger frequencies greater than or equal to the first trigger frequency threshold, and set them as the high-frequency water pump housing defects; Based on the trigger frequencies of the first defect attribute until the Qth defect attribute, extract the defects with attribute trigger frequencies greater than or equal to the second trigger frequency threshold and less than the first trigger frequency threshold, and set them as the low-frequency water pump housing defects; Based on the trigger frequencies of the first defect attribute until the Qth defect attribute, extract the defects with attribute trigger frequencies less than the second trigger frequency threshold, and set them as the water pump housing defects not participating in the detection.

[0007] Among them, calculating the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix includes: Configure the first attribute deviation threshold until the Nth attribute deviation threshold through the client; Based on the first attribute deviation threshold until the Nth attribute deviation threshold, combined with the water pump housing historical samples, count the simultaneous trigger frequency of the defect and the first attribute when only the first attribute is in the deviation state and the remaining attributes are in the non-deviation state, and set it as the first attribute defect support degree; Until the Nth attribute defect support degree is obtained; Calculate the ratio of the sum of the first attribute defect support degree until the Nth attribute defect support degree to the defect support degree summation to obtain the weight distribution vector of the first attribute until the Nth attribute; Based on the weight distribution vector, the production control parameter deviation vector matrix and the sample production control parameter deviation vector matrix are weighted respectively to obtain a production control parameter deviation vector update matrix and a sample production control parameter deviation vector update matrix; Calculate the matrix similarity between the production control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix.

[0008] Among them, the trigger frequency threshold includes a first type of first trigger frequency threshold and a first type of second trigger frequency threshold, and the first type of first trigger frequency threshold is greater than the first type of second trigger frequency threshold; the trigger frequency threshold includes a second type of first trigger frequency threshold and a second type of second trigger frequency threshold, and the second type of first trigger frequency threshold is greater than the second type of second trigger frequency threshold; the first type of first trigger frequency threshold is greater than the second type of first trigger frequency threshold, including: When the defect attribute belongs to a failure type defect, use the second type of first trigger frequency threshold and the second type of second trigger frequency threshold; When the defect belongs to a non-failure type defect, use the first type of first trigger frequency threshold and the first type of second trigger frequency threshold.

[0009] Optionally, retrieve the low-frequency feature extractor to perform feature extraction on the water pump housing images that meet a set proportion of the production control parameter deviation vector matrix, including: Calculate the ratio of the number of appearance samples of the trigger frequency of the low-frequency water pump housing defect to the total number of historical samples of the water pump housing, and set it as the sampling rate threshold; Using the sampling rate threshold as the minimum sampling rate, retrieve the low-frequency feature extractor to perform feature extraction on the water pump housing images that meet a set proportion of the production control parameter deviation vector matrix.

[0010] Optionally, retrieve the high-frequency feature extractor to perform feature extraction on all the water pump housing images that meet the production control parameter deviation vector matrix, retrieve the low-frequency feature extractor to perform feature extraction on the water pump housing images that meet a set proportion of the production control parameter deviation vector matrix, and obtain the image feature extraction result, further including: Statistically calculate the historical average computing power of the high-frequency feature extractor and the low-frequency feature extractor respectively to obtain the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power; Based on the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power, extract the first high-frequency feature extractor and the first low-frequency feature extractor whose required computing power is greater than or equal to the computing power configuration threshold, and perform cloud computing power scheduling configuration; Based on the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power, a second high-frequency feature extractor and a second low-frequency feature extractor whose required computing power is less than the computing power configuration threshold are extracted to perform edge computing power scheduling configuration.

[0011] In a second aspect, the present invention provides a water pump casing image processing system for quality inspection, comprising: A deviation vector matrix extraction unit is used to compare the production expected control parameters and production monitoring control parameters of the water pump housing to obtain the production control parameter deviation vector matrix; A defect type identification unit, used to retrieve high-frequency water pump casing defects and low-frequency water pump casing defects that satisfy the production control parameter deviation vector matrix and whose trigger frequency is greater than or equal to a trigger frequency threshold from historical samples of water pump casings; A defect feature processing unit, used to process the high-frequency water pump housing defect and the low-frequency water pump housing defect based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor; The image feature extraction unit is used to call the high-frequency feature extractor to perform feature extraction on all water pump casing images that satisfy the production control parameter deviation vector matrix, call the low-frequency feature extractor to perform feature extraction on a set proportion of water pump casing images that satisfy the production control parameter deviation vector matrix, and obtain image feature extraction results.

[0012] By implementing the present invention, it is possible to compare the production expected control parameters and production monitoring control parameters of the water pump housing, obtain the production control parameter deviation vector matrix, and accurately measure the deviation value between the ideal production parameters and the actual production parameters; By implementing the present invention, it is possible to retrieve high-frequency pump casing defects and low-frequency pump casing defects that meet the production control parameter deviation vector matrix and whose trigger frequency is greater than or equal to the trigger frequency threshold from historical samples of the pump casing, set the trigger frequency threshold in a targeted manner according to different defects, reasonably feedback quality information, and improve the rationality and scientificity of the inspection sample data; By implementing the present invention, the high-frequency water pump housing defect and the low-frequency water pump housing defect can be processed based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor, thereby realizing targeted extraction of water pump housing defects with different occurrence frequencies, thereby saving inspection resources and improving extraction efficiency; By implementing the present invention, it is possible to call the high-frequency feature extractor to extract features from all the pump housing images that meet the production control parameter deviation vector matrix, and call the low-frequency feature extractor to extract features from the pump housing images with a set proportion that meet the production control parameter deviation vector matrix, so as to obtain the image feature extraction result, so as to accurately extract and feedback various types of defect feature images according to the inspection requirements under limited hardware conditions, and maximize the inspection effect.

[0013] In summary, by implementing the present invention, it is possible to predict the occurrence probability of different housing defect types and evaluate their importance through production parameters (such as casting pressure, mold temperature, material batch), dynamically adjust the detection resources, so as to achieve the technical effects of saving inspection resources and improving inspection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic flow chart of a method for processing pump housing images for quality inspection provided by the present invention; Figure 2 It is a schematic structural diagram of a system for processing pump housing images for quality inspection provided by the present invention.

[0015] In the drawings, the components represented by each reference numeral are as follows: Deviation vector matrix extraction unit 11, defect type recognition unit 12, defect feature processing unit 13, image feature extraction unit 14. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0017] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0018] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to implement and use the present invention. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0019] Embodiment 1, as Figure 1 shown, an embodiment of the present invention provides an image processing method for a water pump housing for quality inspection, including: S100: Comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix; S200: Retrieving from the historical samples of the water pump housing high-frequency water pump housing defects and low-frequency water pump housing defects that satisfy the production control parameter deviation vector matrix and whose triggering frequency is greater than or equal to the triggering frequency threshold; S300: Processing the high-frequency water pump housing defects and the low-frequency water pump housing defects based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor; S400: Invoking the high-frequency feature extractor to perform feature extraction on all water pump housing images that satisfy the production control parameter deviation vector matrix, and invoking the low-frequency feature extractor to perform feature extraction on a set proportion of water pump housing images that satisfy the production control parameter deviation vector matrix to obtain an image feature extraction result.

[0020] Among them, in step S100 of the embodiment of the present application, comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix includes: Performing normalization processing on the production expected control parameters and the production monitoring control parameters respectively to obtain a production expected control eigenvalue and a production monitoring control eigenvalue; Performing simultaneous time-domain and same-attribute absolute deviation value calculations on the production expected control eigenvalue and the production monitoring control eigenvalue to obtain first-attribute deviation time-series information to Nth-attribute deviation time-series information, where the time series covers the entire production time zone and the matrix elements are configured as empty when the attribute is not set at the corresponding moment.

[0021] In an embodiment of the present application, a production control parameter deviation vector matrix is ​​established in order to use it as a basis for screening selected historical samples of water pump casings, and to retrieve and classify high-frequency water pump casing defects and low-frequency water pump casing defects. Specifically, to establish a production control parameter deviation vector matrix, it is first necessary to calculate the actual deviations between the production expected control parameters and the production monitoring control parameters. Among them, the production expected control parameters refer to the preset parameters in the water pump casing production process, that is, the parameters expected to be achieved, but in the actual production process, due to various reasons, the predetermined production expected control parameters are often not achieved. Therefore, it is necessary to monitor the actual parameters in the trial production process as production monitoring control parameters.

[0022] Exemplarily, the production expected control parameters and production monitoring control parameters include but are not limited to casting pressure, mold temperature, cooling rate, etc. For example, if in the production expected control parameters, the casting pressure in a certain time domain is set to 60MPa, the mold temperature is set to 230°C, and the cooling rate is set to 70°C / h, while in the production monitoring control parameters, the actual casting pressure may be 62MPa, the mold temperature may be 227°C, and the cooling rate may be 60°C / h.

[0023] In the embodiment of the present application, in order to facilitate calculation, it is necessary to normalize the production expectation control parameter and the production monitoring control parameter respectively. Exemplarily, the maximum-minimum value normalization method can be adopted, that is, the normalized value = (parameter to be normalized - upper limit of parameter) / (upper limit of parameter - lower limit of parameter). Among them, the upper limit of parameter and the lower limit of parameter are the maximum and minimum values ​​that the parameter to be normalized can reach under normal circumstances, which can be summarized from all production and processing data within a historical time (such as 3 months). If the range of all production and processing data of the casting pressure in the historical time is 50-70MPa, the production expectation control value is 60MPa, and the production monitoring control value is 62MPa, then the normalized value of the production expectation control value is (60-50) / (70-50)=0.5, and the normalized value of the production monitoring control value is (62-50) / (70-50)=0.6. By this method, the production expectation control parameters and production monitoring control parameters of different parameter types such as casting pressure, mold temperature, cooling rate, etc. can be normalized respectively, and the normalized values ​​obtained are the production expectation control characteristic values ​​and production monitoring control characteristic values. For example, if only the three characteristics of casting pressure, mold temperature, and cooling rate are calculated, the production expectation control characteristic values ​​and production monitoring control characteristic values ​​can be three-dimensional characteristic vectors in the format of (0.6, 0.3, 0.8) and (0.4, 0.4, 0.6).

[0024] Exemplarily, the calculation method of the absolute deviation value of the production expected control feature value and the production monitoring control feature value can be: absolute deviation value = |production expected control feature value - production monitoring control feature value|. For example, at a certain moment, when the production expected control feature values and production monitoring control feature values corresponding to the casting pressure, mold temperature, and cooling rate are (0.6, 0.3, 0.8) and (0.4, 0.4, 0.6) respectively, the absolute deviation value of the production expected control feature value and production monitoring control feature value of the casting pressure can be |0.6 - 0.4| = 0.2, the absolute deviation value of the production expected control feature value and production monitoring control feature value of the mold temperature can be |0.3 - 0.4| = 0.1, and the absolute deviation value of the production expected control feature value and production monitoring control feature value of the cooling rate can be |0.8 - 0.6| = 0.2. Then, the corresponding production control parameter deviation vector is (0.2, 0.1, 0.2). By collecting the production control parameter deviation vectors at multiple time points, the production control parameter deviation vector matrix can be obtained. Through the above method, the deviation time series information corresponding to each parameter of the same attribute at each moment can be calculated, that is, the first attribute deviation time series information to the Nth attribute deviation time series information. Wherein, N represents the number of production expected control parameters and production monitoring control parameters of different attributes. If only the values of the casting pressure, mold temperature, and cooling rate are monitored, then N is 3. The acquisition frequency of the time series parameter information can be determined according to actual needs. For example, the production expected control parameters and production monitoring control parameters can be collected once every 10 minutes. The total parameter acquisition time range (i.e., the time series coverage time domain) is the entire production process of the pump housing (i.e., the full production time zone). For example, this production process can be 10 hours. If a parameter of a certain attribute, such as the casting pressure, is no longer set after casting in the actual production process, the position of the feature value corresponding to the parameter of this attribute can be in a vacant state.

[0025] In step S200 of this embodiment, from the historical samples of the water pump housing, high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and whose trigger frequency is greater than or equal to the trigger frequency threshold are retrieved. The trigger frequency threshold includes a first trigger frequency threshold and a second trigger frequency threshold, and the first trigger frequency threshold is greater than the second trigger frequency threshold, including: Extract the sample production control parameter deviation vector matrix from the historical samples of the water pump housing; Calculate the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix. If it is greater than or equal to the matrix similarity threshold, add it to the selected historical samples of the water pump housing. Otherwise, update the historical samples of the water pump housing; When the number of historical samples of the selected water pump housing is greater than or equal to the set number, the trigger frequencies of the first defect attribute to the Qth defect attribute are counted, where the set number ≥ 10000, and all historical samples belong to the set time window; Based on the trigger frequencies of the first defect attribute to the Qth defect attribute, defects with attribute trigger frequencies greater than or equal to the first trigger frequency threshold are extracted and set as the high-frequency water pump housing defects; Based on the trigger frequencies of the first defect attribute to the Qth defect attribute, defects with attribute trigger frequencies greater than or equal to the second trigger frequency threshold and less than the first trigger frequency threshold are extracted and set as the low-frequency water pump housing defects; Based on the trigger frequencies of the first defect attribute to the Qth defect attribute, defects with attribute trigger frequencies less than the second trigger frequency threshold are extracted and set as water pump housing defects not participating in detection.

[0026] In the embodiments of the present application, the sample production control parameter deviation vector matrix for distinguishing high-frequency water pump housing defects and low-frequency water pump housing defects can be obtained by the method described in S100. Then, it is necessary to collect the production expected control parameters and the production monitoring control parameter set for water pump housing production within the historical time (such as 3 months), extract the sample production control parameter deviation vector matrix from them, calculate multiple sample production control parameter deviation vector matrices, and use them as the historical samples of the water pump housing.

[0027] Among them, calculating the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix includes: Through the user terminal, configure the first attribute deviation threshold to the Nth attribute deviation threshold; Based on the first attribute deviation threshold to the Nth attribute deviation threshold, combined with the historical samples of the water pump housing, count the simultaneous trigger frequency of the defect and the first attribute when only the first attribute is in the deviation state and the remaining attributes are in the non-deviation state, and set it as the first attribute defect support degree; Until the Nth attribute defect support degree is obtained; Calculate the ratio of the sum of the first attribute defect support degree to the Nth attribute defect support degree to the sum of the defect support degrees to obtain the weight distribution vector of the first attribute to the Nth attribute; Based on the weight distribution vector, weight the production control parameter deviation vector matrix and the sample production control parameter deviation vector matrix respectively to obtain the production control parameter deviation vector updated matrix and the sample production control parameter deviation vector updated matrix; Calculate the matrix similarity between the production control parameter deviation vector updated matrix and the sample production control parameter deviation vector updated matrix.

[0028] In the embodiment of the present application, the first attribute deviation threshold to the Nth attribute deviation threshold are the criteria for judging whether the parameters of the attribute are in a deviation state, and these thresholds can be set according to actual production needs. For example, the absolute deviation value threshold of the production expected control characteristic value and the production monitoring control characteristic value of the mold temperature can be set to 0.2. When the actual deviation value is greater than 0.2, it can be judged that the deviation state of the attribute (mold temperature) is deviation. If the actual deviation value is less than this threshold (0.2), it is judged that the deviation state of the attribute is not deviation. In the same way, the first attribute deviation threshold to the Nth attribute deviation threshold can be determined, and whether the corresponding attribute is in a deviation state can be judged.

[0029] In the embodiment of the present application, when the first attribute is in a deviation state and the remaining attributes are in a non-deviation state, the simultaneous triggering frequency of the defect and the first attribute specifically refers to the frequency in historical data that only when the deviation state of the first attribute (such as mold temperature) is deviation and the deviation states of the remaining attributes (such as casting pressure, etc.) are non-deviation, a defective water pump housing is finally produced. For example, within the data of historical time (such as 3 months), when the deviation state of the first attribute (such as mold temperature) is deviation and the deviation states of the remaining attributes (such as casting pressure, etc.) are non-deviation, the number of produced water pump housings is 100, and among them, the number of defective (such as having air holes, shrinkage porosity, etc.) water pump housings is 10. Then the value of the simultaneous triggering frequency of the defect and the first attribute is 10 / 100 = 0.1 = 10%. Then this value can be set as the defect support degree of the first attribute to represent the correlation degree between this attribute and the water pump housing defect.

[0030] Through the above method, the defect support degree of the first attribute, the defect support degree of the second attribute, and until the defect support degree of the Nth attribute can be calculated. Where N is the number of parameters of different attributes (such as casting pressure, mold temperature, cooling rate).

[0031] After obtaining the defect support degrees of the first attribute to the defect support degree of the Nth attribute, the distribution vectors of the first attribute to the Nth attribute can be weighted according to the defect support degrees to obtain the weight distribution vectors of the first attribute to the Nth attribute. Exemplarily, if there are three attributes: casting pressure, mold temperature, and cooling rate, and assuming that the defect support degrees of the three attributes calculated through the above steps are 10%, 20%, and 20% in sequence, then the weighted values of the distribution vectors calculated according to the defect support degrees of the above three attributes are 10% / (10% + 20% + 20%) = 0.2, 20% / (10% + 20% + 30%) = 0.4, 20% / (10% + 20% + 30%) = 0.4 respectively. It is not difficult to see that the sum of the weighted values is 1. After obtaining the weighted values of each attribute, it is necessary to weight the production control parameter deviation vector matrix and the sample production control parameter deviation vector matrix. Through the above calculation method, the weight distribution vectors of the first attribute to the Nth attribute can be calculated. Where N is the number of parameters of different attributes (such as casting pressure, mold temperature, cooling rate).

[0032] Exemplarily, assuming there are three parameters: casting pressure, mold temperature, and cooling rate, if a certain deviation vector in the production control parameter deviation vector matrix or the sample production control parameter deviation vector matrix is (0.6, 0.3, 0.8), then it is necessary to multiply the corresponding values of the three parameters in the vector by the weighted values of the weight distribution vector calculated above. For example, when the weighted values are 0.2, 0.4, 0.4 respectively, the weighted vector is. By weighting all the vectors in the deviation vector matrix, the deviation vector update matrix can be obtained. Through this method, the production control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix can be calculated.

[0033] Finally, it is necessary to calculate the matrix similarity between the production control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix. Among them, when monitoring the data of three parameters: casting pressure, mold temperature, and cooling rate, both of the above two matrices are matrices with 3 columns and M rows, where 3 represents three monitoring parameters, and M is the number of sampling times of the above three parameter data in the previous full production time zone (such as 20 times).

[0034] The specific matrix similarity comparison method can be the method of combining dynamic time warping (DTW) with multi-dimensional parameter weighted average. The specific steps are to calculate the DTW distance for each parameter column (casting pressure, mold temperature, cooling rate) of the matrix to be compared separately, and then synthesize according to the weights, which is applicable to the scenario where the physical meanings of each parameter are independent and different weights are required. For example, the DTW distance Dtotal = 0.4D1 (casting pressure DTW distance) 0.3D2 (mold temperature DTW distance) 0.3D3 (cooling rate DTW distance). Then map the DTW distance Dtotal to the interval of 0 to 1 to obtain the similarity value. For example, define similarity = 1 / (1 + Dtotal). The closer the similarity value is to 1, the more similar the control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix are.

[0035] Furthermore, a matrix similarity threshold can be set, such as 0.8. When the similarity between the control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix is greater than this threshold, it indicates that the similarity between the two is extremely high, and the sample production control parameter deviation vector update matrix can be used as the selected historical sample of the water pump housing; when it is less than this threshold, it indicates that the similarity between the two is not high, and the historical sample of the water pump housing needs to be updated so as to continue to select more samples with a similarity greater than or equal to this threshold (such as 0.8) from the updated historical samples as the selected historical sample of the water pump housing. When the number of selected historical samples of the water pump housing is greater than or equal to the set quantity (in this embodiment of the application, this quantity can be ≥ 10,000), count the first defect attribute trigger frequency until the Qth defect attribute trigger frequency, the first defect attribute trigger frequency until the Qth defect attribute trigger frequency.

[0036] Among them, Q represents the number of types of water pump housing defects detected during the inspection of the selected historical samples of the water pump housing. For example, during the inspection of the selected historical samples of the water pump housing, five types of defects such as air holes, cracks, shrinkage porosity, burrs, and pits have been detected. At this time, the value of Q is 5. The defect attribute trigger frequency represents the ratio of the number of samples with a certain defect in the selected historical samples of the water pump housing to the total number of selected historical samples of the water pump housing. For example, if the total number of selected historical samples of the water pump housing is 10,000 and the number of historical samples of the water pump housing with burrs found during the inspection is 6,000, the trigger frequency of this defect attribute (burrs) is 6,000 / 10,000 = 0.6 = 60%. Through the above method, the first defect attribute trigger frequency until the Qth defect attribute trigger frequency can be calculated.

[0037] Among them, the trigger frequency threshold includes a first type of first trigger frequency threshold and a first type of second trigger frequency threshold, and the first type of first trigger frequency threshold is greater than the first type of second trigger frequency threshold; The trigger frequency threshold includes a second type of first trigger frequency threshold and a second type of second trigger frequency threshold, and the second type of first trigger frequency threshold is greater than the second type of second trigger frequency threshold; The first type of first trigger frequency threshold being greater than the second type of first trigger frequency threshold includes: When the defect attribute belongs to the failure type of defect, use the second type of first trigger frequency threshold and the second type of second trigger frequency threshold; When the defect belongs to a non-failure type defect, the first type of first trigger frequency threshold and the first type of second trigger frequency threshold are used.

[0038] In the embodiments of the present application, since different image processing strategies need to be adopted for different defect attributes, it is necessary to divide the defect attributes. Among them, failure type defects refer to relatively serious defects that may cause the failure of the water pump housing, including cracks, shrinkage porosity, etc. Non-failure type defects are defects with a relatively low degree that do not cause the failure of the water pump housing, including burrs, potholes, etc. The division of failure type defects and non-failure type defects can be to first count all Q-type defect attributes detected in the selected historical samples of the water pump housing, and then technicians analyze according to the previous causes of water pump housing failure to evaluate the possibility of each defect causing the failure of the water pump housing during use. Based on this, all Q-type defect attributes are divided into failure type defects and non-failure type defects, and a defect classification index table is established according to the division results. By inputting the defect name, the defect category can be output. This analysis method and the method for establishing the classification index table are prior arts and will not be elaborated here.

[0039] Further, the second type of first trigger frequency threshold and the second type of second trigger frequency threshold configured for relatively serious failure type defects should be smaller so as to be able to detect these serious defects in a timely manner. For example, the second type of first trigger frequency threshold can be set to 40% and the second type of second trigger frequency threshold can be set to 30%.

[0040] For non-failure type defects with a relatively low degree that do not cause the failure of the water pump housing, the first type of first trigger frequency threshold and the first type of second trigger frequency threshold should be smaller. For example, the first type of first trigger frequency threshold can be set to 60% and the first type of second trigger frequency threshold can be set to 50%. To reduce unnecessary defect recognition. And, the first type of first trigger frequency threshold is greater than the second type of first trigger frequency threshold.

[0041] Then, defects with an attribute trigger frequency greater than or equal to the first trigger frequency threshold (such as 40%) are respectively extracted and set as the high-frequency water pump housing defects. Based on the first defect attribute trigger frequency until the Qth defect attribute trigger frequency, defects with an attribute trigger frequency greater than or equal to the second trigger frequency threshold (such as 20%) and less than the first trigger frequency threshold (such as 40%) are extracted and set as the low-frequency water pump housing defects. So as to formulate different inspection strategies according to low-frequency and high-frequency defects. Among them, based on the first defect attribute trigger frequency until the Qth defect attribute trigger frequency, defects with an attribute trigger frequency less than the second trigger frequency threshold (20%) are extracted and set as water pump housing defects that do not participate in detection.

[0042] In step S300 of the embodiment of the present application, based on a predefined defect feature extractor calibration table, the high-frequency water pump housing defects and the low-frequency water pump housing defects are processed to obtain a high-frequency feature extractor and a low-frequency feature extractor.

[0043] The defect feature extractor calibration table contains the feature information of all Q-class water pump housing defects determined by the above method, including which type of defect this defect belongs to, whether it is a high-frequency water pump housing defect, a low-frequency water pump housing defect, or a defect that does not participate in the detection of the water pump housing, and which type of defect it belongs to, whether it is a failure type defect or a non-failure type defect, and so on. Then, according to the feature information of a specific defect, the corresponding high-frequency feature extractor or low-frequency feature extractor is matched. The feature extraction logics of the high-frequency feature extractor and the low-frequency feature extractor are different. For the defect types matched to the high-frequency feature extractor, the inspection strategy is to extract the image feature information of all samples to prevent missing high-frequency water pump housing defects; for the defect types matched to the low-frequency feature extractor, the inspection strategy is to extract the image feature information of some samples for sampling inspection to save inspection resources. The specific sampling inspection method is as described in step S400.

[0044] The high-frequency feature extractor and the low-frequency feature extractor can identify the image of the water pump housing, identify defects such as pores, cracks, shrinkage porosity, burrs, and potholes contained in the water pump housing image, and submit them to technicians for review.

[0045] The high-frequency feature extractor and the low-frequency feature extractor have the same structure, and the only difference lies in the training data, which are high-frequency defect types and low-frequency defect types. An improved U-Net architecture based on multi-scale feature fusion can be used, and an attention mechanism and a data augmentation strategy are introduced. Specifically, ResNet-50 is used as the encoder, and its residual structure can effectively extract deep features and avoid gradient disappearance. The decoder part adopts a U-Net symmetric structure. And several improvement modules are set, such as a multi-scale feature pyramid (FPN), which is added after the output of the second to fifth stages of the encoder, and the channels are unified through 1x1 convolution and then upsampled and fused to enhance the sensitivity to tiny defects (such as burrs); and a channel attention module (SE Block), which is introduced before each layer of skip connection in the decoder to dynamically adjust the feature channel weights and suppress background interference; using deformable convolution (DeformableConv) to replace the 3x3 convolution in the third stage of the backbone network to adapt to irregular shapes such as cracks and shrinkage porosity.

[0046] The optimizer of the feature extractor can use AdamW, with the initial learning rate set to 0.001 and the weight decay set to 0.01. The loss function uses Focal Loss + Dice, where α = 0.8 and γ = 2.0 are set to alleviate class imbalance (high proportion of pores). The batch size is set to 16. During learning, the cosine annealing method is used for learning rate scheduling, with a cycle of 20 epochs and a minimum learning rate of 0.0001; training is terminated if the mAP of the validation set does not improve for 10 consecutive epochs; the gradient clipping threshold is set to 5.0 to prevent gradient explosion.

[0047] The training data are pictures of various defects that occurred within the historical time. The pictures of each type of defect are pre-annotated with the defect type in advance, and the number of training pictures for each type of defect is not less than 500. They are divided into a validation set and a training set according to a ratio of 2:8. According to the categories of the aforementioned high-frequency water pump housing defects and low-frequency water pump housing defects, they are divided into training data for the high-frequency feature extractor and training data for the low-frequency feature extractor, which are used to train the high-frequency feature extractor and the low-frequency feature extractor respectively.

[0048] During training, pre-training is first performed, with the initial learning rate set to 0.001 and training for 50 epochs; the model evaluation metric can be the mAP of the validation set, calculated using the COCO standard, with the IOU threshold range [0.5:0.95], weighted and averaged according to the defect type. It can be set that when this value ≥ 0.92, the model converges, and the high-frequency feature extractor and the low-frequency feature extractor are obtained. When inputting the water pump housing image, the corresponding low-frequency defect features and high-frequency defect features can be extracted.

[0049] Among them, the water pump housing image contains complete images of all surfaces of the water pump housing, which can be obtained through monitoring devices at the production site. Such monitoring devices are prior art and will not be elaborated here.

[0050] In step S400 of the embodiment of the present application, the low-frequency feature extractor is called to perform feature extraction on the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix, including: Calculate the ratio of the number of occurrence samples of the trigger frequency of the low-frequency water pump housing defect to the total number of historical samples of the water pump housing, and set it as the sampling rate threshold; Taking the sampling rate threshold as the minimum sampling rate, the low-frequency feature extractor is called to perform feature extraction on the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix.

[0051] In one embodiment, the number of occurrence samples (such as 2000) of the trigger frequency of a certain low-frequency water pump housing defect (such as surface pitting) can be calculated, and the ratio of this number to the total number of historical samples of the selected water pump housing (such as 10000) is set as the sampling rate threshold. That is, the sampling ratio for this low-frequency water pump housing defect during the sampling process is not less than 0.2 (20%) of the total production quantity. Using the trigger frequency of the low-frequency water pump housing defect as the sampling rate threshold can ensure that more defects can be found as much as possible while saving inspection resources during the sampling process. Then, it is necessary to retrieve the low-frequency feature extractor for extracting the corresponding low-frequency feature images, and perform feature extraction on the water pump housing images that meet a set ratio of the production control parameter deviation vector matrix (this set ratio is greater than or equal to the trigger frequency of the low-frequency water pump housing defect, such as greater than or equal to 20%).

[0052] In step S400 of the embodiment of the present application, retrieve the high-frequency feature extractor, perform feature extraction on all water pump housing images that meet the production control parameter deviation vector matrix, retrieve the low-frequency feature extractor, perform feature extraction on the water pump housing images that meet a set ratio of the production control parameter deviation vector matrix, and obtain the image feature extraction result. It further includes: Respectively count the historical average computing power of the high-frequency feature extractor and the low-frequency feature extractor to obtain the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power; Based on the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power, extract the first high-frequency feature extractor and the first low-frequency feature extractor whose required computing power is greater than or equal to the computing power configuration threshold, and perform cloud computing power scheduling configuration; Based on the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power, extract the second high-frequency feature extractor and the second low-frequency feature extractor whose required computing power is less than the computing power configuration threshold, and perform edge computing power scheduling configuration.

[0053] In the embodiment of the present application, for the purpose of improving the inspection efficiency, it is also necessary to optimize the computing power configuration of the high-frequency feature extractor and the low-frequency feature extractor. The optimization method is to statistically analyze the historical computing power requirements of the high-frequency feature extractor and the low-frequency feature extractor according to historical data, and perform computing power scheduling according to the computing power requirements to support the work of the high-frequency feature extractor and the low-frequency feature extractor and ensure the inspection efficiency.

[0054] First, it is necessary to obtain the average computing power consumed by the high-frequency feature extractor and the low-frequency feature extractor when performing feature extraction tasks within a historical time period (such as in the most recent 100 feature extraction tasks), and then compare this average computing power with the computing power configuration threshold to make a computing power scheduling strategy accordingly.

[0055] Among them, the computing power configuration threshold is the maximum computing power that the terminal system can provide (such as 2 TFLOPS), which depends on the hardware facilities of the terminal system. When the computing power required for high-frequency feature extraction and the computing power required for low-frequency feature extraction (such as 3 TFLOPS) are greater than this threshold (such as 2 TFLOPS), it means that the terminal system cannot provide sufficient computing power support for the computing power required for high-frequency feature extraction and low-frequency feature extraction. To ensure the normal progress of feature extraction, it is necessary to schedule additional cloud computing power for the first high-frequency feature extractor and the first low-frequency feature extractor to support the normal operation of the first high-frequency feature extractor and the first low-frequency feature extractor. Among them, the cloud computing power can be obtained through the computing power leasing service provided by a third-party cloud platform, or can be provided by remote computing units deployed in other systems, which will not be elaborated here.

[0056] When the computing power required for high-frequency feature extraction and the computing power required for low-frequency feature extraction are less than the computing power configuration threshold, it means that the terminal system still has redundant computing power reserves, which can provide sufficient computing power support for the computing power required for high-frequency feature extraction and low-frequency feature extraction. Only the redundant computing power needs to be scheduled inside the terminal system to support the first high-frequency feature extractor or the first low-frequency feature extractor that is performing operations, and the edge computing power scheduling configuration can be carried out. The specific edge computing power scheduling configuration method is a prior art and will not be elaborated here.

[0057] Embodiment 2, as Figure 2 shown, based on the same inventive concept as the method for processing images of a water pump housing for quality inspection provided in Embodiment 1, the embodiment of the present invention also provides a system for processing images of a water pump housing for quality inspection, including: A deviation vector matrix extraction unit 11, configured to compare the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix; A defect type recognition unit 12, configured to retrieve from the historical samples of the water pump housing high-frequency water pump housing defects and low-frequency water pump housing defects that satisfy the production control parameter deviation vector matrix and the triggering frequency is greater than or equal to the triggering frequency threshold; A defect feature processing unit 13, configured to process the high-frequency water pump housing defects and the low-frequency water pump housing defects based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor; An image feature extraction unit 14, configured to retrieve the high-frequency feature extractor to perform feature extraction on all water pump housing images that satisfy the production control parameter deviation vector matrix, retrieve the low-frequency feature extractor to perform feature extraction on a set proportion of water pump housing images that satisfy the production control parameter deviation vector matrix, and obtain an image feature extraction result.

[0058] Further, the deviation vector matrix extraction unit 11 is further configured to: Normalize the production expected control parameters and the production monitoring control parameters respectively to obtain the production expected control eigenvalue and the production monitoring control eigenvalue; Calculate the absolute deviation values of the same time domain and the same attribute for the production expected control eigenvalue and the production monitoring control eigenvalue to obtain the first attribute deviation time series information to the Nth attribute deviation time series information, where the time series covers the entire production time zone, and the matrix elements are configured as empty when the attribute is not set at the corresponding moment.

[0059] Furthermore, the defect type identification unit 12 is also used for: Extract the sample production control parameter deviation vector matrix from the historical samples of the water pump housing; Calculate the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix. If it is greater than or equal to the matrix similarity threshold, add it to the selected historical samples of the water pump housing. Otherwise, update the historical samples of the water pump housing; When the number of the selected historical samples of the water pump housing is greater than or equal to the set number, count the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, where the set number ≥ 10000, and all historical samples belong to the set time window; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extract the defects whose attribute trigger frequency is greater than or equal to the first trigger frequency threshold, and set them as the high-frequency water pump housing defects; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extract the defects whose attribute trigger frequency is greater than or equal to the second trigger frequency threshold and less than the first trigger frequency threshold, and set them as the low-frequency water pump housing defects; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extract the defects whose attribute trigger frequency is less than the second trigger frequency threshold, and set them as the water pump housing defects not participating in the detection.

[0060] Among them, calculating the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix includes: Configure the first attribute deviation threshold to the Nth attribute deviation threshold through the user terminal; Based on the first attribute deviation threshold to the Nth attribute deviation threshold, combined with the historical samples of the water pump housing, count the simultaneous trigger frequency of the defect and the first attribute when only the first attribute is in the deviation state and the other attributes are in the non-deviation state, and set it as the first attribute defect support degree; Until the Nth attribute defect support degree is obtained; Calculate the ratio of the support degree of the first attribute defect to the sum of the support degrees of the Nth attribute defect until the Nth attribute defect support degree, to obtain the weight distribution vector of the first attribute to the Nth attribute; Based on the weight distribution vector, weight the production control parameter deviation vector matrix and the sample production control parameter deviation vector matrix respectively, to obtain the production control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix; Calculate the matrix similarity of the production control parameter deviation vector update matrix and the sample production control parameter deviation vector update matrix.

[0061] The trigger frequency threshold includes a first type of first trigger frequency threshold and a first type of second trigger frequency threshold, and the first type of first trigger frequency threshold is greater than the first type of second trigger frequency threshold; the trigger frequency threshold includes a second type of first trigger frequency threshold and a second type of second trigger frequency threshold, and the second type of first trigger frequency threshold is greater than the second type of second trigger frequency threshold; the first type of first trigger frequency threshold is greater than the second type of first trigger frequency threshold, including: When the defect attribute belongs to the failure type defect, use the second type of first trigger frequency threshold and the second type of second trigger frequency threshold; When the defect belongs to the non-failure type defect, use the first type of first trigger frequency threshold and the first type of second trigger frequency threshold.

[0062] Furthermore, the image feature extraction unit 14 is also used for: Retrieve the low-frequency feature extractor, and perform feature extraction on the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix, including: Calculate the ratio of the number of appearance samples of the trigger frequency of the low-frequency water pump housing defect to the total number of historical samples of the water pump housing, and set it as the sampling rate threshold; Taking the sampling rate threshold as the minimum sampling rate, retrieve the low-frequency feature extractor, and perform feature extraction on the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix.

[0063] Take the high-frequency feature extractor, perform feature extraction on all the water pump housing images that meet the production control parameter deviation vector matrix, retrieve the low-frequency feature extractor, and perform feature extraction on the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix, to obtain the image feature extraction result, and further include: Respectively count the historical average computing power of the high-frequency feature extractor and the low-frequency feature extractor, to obtain the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power; Based on the computing power requirements for high-frequency feature extraction and the computing power requirements for low-frequency feature extraction, extract the first high-frequency feature extractor and the first low-frequency feature extractor whose required computing power is greater than or equal to the computing power configuration threshold, and perform cloud computing power scheduling configuration; Based on the computing power requirements for high-frequency feature extraction and the computing power requirements for low-frequency feature extraction, extract the second high-frequency feature extractor and the second low-frequency feature extractor whose required computing power is less than the computing power configuration threshold, and perform edge computing power scheduling configuration.

[0064] It should be noted that in the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0065] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0067] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus, causing a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process, such that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps for implementing the functions specified in one block or a plurality of blocks.

[0069] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept.

[0070] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An image processing method for a water pump housing used in quality inspection, characterized in that, Including: Comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix; Retrieving from the historical samples of the water pump housing high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and have a trigger frequency greater than or equal to the trigger frequency threshold; Based on a predefined defect feature extractor calibration table, processing the high-frequency water pump housing defects and the low-frequency water pump housing defects to obtain a high-frequency feature extractor and a low-frequency feature extractor; Invoking the high-frequency feature extractor to perform feature extraction on all water pump housing images that meet the production control parameter deviation vector matrix, and invoking the low-frequency feature extractor to perform feature extraction on a set proportion of water pump housing images that meet the production control parameter deviation vector matrix to obtain an image feature extraction result.

2. The method according to claim 1, wherein Comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix, including: Performing normalization processing on the production expected control parameters and the production monitoring control parameters respectively to obtain a production expected control eigenvalue and a production monitoring control eigenvalue; Calculating the absolute deviation values of the same time domain and the same attribute for the production expected control eigenvalue and the production monitoring control eigenvalue to obtain the first attribute deviation time series information to the Nth attribute deviation time series information, where the time series covers the entire production time zone and the matrix elements are configured as empty when the attribute is not set at the corresponding moment.

3. The method according to claim 2, wherein Retrieving from the historical samples of the water pump housing high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and have a trigger frequency greater than or equal to the trigger frequency threshold, where the trigger frequency threshold includes a first trigger frequency threshold and a second trigger frequency threshold, and the first trigger frequency threshold is greater than the second trigger frequency threshold, including: Extracting a sample production control parameter deviation vector matrix from the historical samples of the water pump housing; Calculating the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix, and if it is greater than or equal to the matrix similarity threshold, adding it to the selected historical samples of the water pump housing, otherwise, updating the historical samples of the water pump housing; When the number of the selected historical samples of the water pump housing is greater than or equal to the set number, counting the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, where the set number ≥ 10000 and all historical samples belong to a set time window; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extracting defects with an attribute trigger frequency greater than or equal to the first trigger frequency threshold and setting them as the high-frequency water pump housing defects; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extracting defects with an attribute trigger frequency greater than or equal to the second trigger frequency threshold and less than the first trigger frequency threshold and setting them as the low-frequency water pump housing defects; Based on the first defect attribute trigger frequency to the Qth defect attribute trigger frequency, extracting defects with an attribute trigger frequency less than the second trigger frequency threshold and setting them as water pump housing defects not participating in detection.

4. The method according to claim 3, wherein Calculating the matrix similarity between the sample production control parameter deviation vector matrix and the production control parameter deviation vector matrix includes: Configuring the first attribute deviation threshold to the Nth attribute deviation threshold through the client; Based on the first attribute deviation threshold to the Nth attribute deviation threshold, combining the historical samples of the water pump housing, counting the simultaneous trigger frequency of the defect and the first attribute when only the first attribute is in the deviation state and the remaining attributes are in the non-deviation state, and setting it as the support degree of the first attribute defect; Until the support degree of the Nth attribute defect is obtained; Calculating the ratio of the support degree of the first attribute defect to the Nth attribute defect to the sum of the support degrees of the defects, and obtaining the weight distribution vector of the first attribute to the Nth attribute; Based on the weight distribution vector, weighting the production control parameter deviation vector matrix and the sample production control parameter deviation vector matrix respectively to obtain the updated matrix of the production control parameter deviation vector and the updated matrix of the sample production control parameter deviation vector; Calculating the matrix similarity between the updated matrix of the production control parameter deviation vector and the updated matrix of the sample production control parameter deviation vector.

5. The method according to claim 3, wherein The trigger frequency threshold includes a first type of first trigger frequency threshold and a first type of second trigger frequency threshold, and the first type of first trigger frequency threshold is greater than the first type of second trigger frequency threshold; the trigger frequency threshold includes a second type of first trigger frequency threshold and a second type of second trigger frequency threshold, and the second type of first trigger frequency threshold is greater than the second type of second trigger frequency threshold; the first type of first trigger frequency threshold is greater than the second type of first trigger frequency threshold, including: When the defect attribute belongs to the failure type defect, use the second type of first trigger frequency threshold and the second type of second trigger frequency threshold; When the defect belongs to the non-failure type defect, use the first type of first trigger frequency threshold and the first type of second trigger frequency threshold.

6. The method according to claim 1, wherein Invoking the low-frequency feature extractor to extract features from the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix, including: Calculating the ratio of the number of appearance samples of the trigger frequency of the low-frequency water pump housing defect to the total number of historical samples of the water pump housing, and setting it as the sampling rate threshold; Taking the sampling rate threshold as the minimum sampling rate, invoking the low-frequency feature extractor to extract features from the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix.

7. The method according to claim 1, characterized in that, Invoking the high-frequency feature extractor to extract features from all the water pump housing images that meet the production control parameter deviation vector matrix, and invoking the low-frequency feature extractor to extract features from the water pump housing images that meet the set ratio of the production control parameter deviation vector matrix, and obtaining the image feature extraction result, further including: Respectively counting the historical average computing power of the high-frequency feature extractor and the low-frequency feature extractor to obtain the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power; Based on the high-frequency feature extraction required computing power and the low-frequency feature extraction required computing power, extracting the first high-frequency feature extractor and the first low-frequency feature extractor whose required computing power is greater than or equal to the computing power configuration threshold for cloud computing power scheduling configuration; Based on the computing power requirements for high-frequency feature extraction and the computing power requirements for low-frequency feature extraction, extract the second high-frequency feature extractor and the second low-frequency feature extractor whose required computing power is less than the computing power configuration threshold, and perform edge computing power scheduling configuration.

8. A water pump housing image processing system for quality inspection, characterized in that, It includes: A deviation vector matrix extraction unit for comparing the production expected control parameters and production monitoring control parameters of the water pump housing to obtain a production control parameter deviation vector matrix; A defect type identification unit for retrieving high-frequency water pump housing defects and low-frequency water pump housing defects that meet the production control parameter deviation vector matrix and whose triggering frequency is greater than or equal to the triggering frequency threshold from the historical samples of the water pump housing; A defect feature processing unit for processing the high-frequency water pump housing defects and the low-frequency water pump housing defects based on a predefined defect feature extractor calibration table to obtain a high-frequency feature extractor and a low-frequency feature extractor; An image feature extraction unit for retrieving the high-frequency feature extractor to extract features from all water pump housing images that meet the production control parameter deviation vector matrix, retrieving the low-frequency feature extractor to extract features from a set proportion of water pump housing images that meet the production control parameter deviation vector matrix, and obtaining an image feature extraction result.

Images