A digital identification method for electromechanical products

Abstract

The application relates to the technical field of electromechanical product inspection, and discloses a kind of electromechanical product digitization identification method, comprising: obtaining the image of electromechanical product, and identifying the state of electromechanical product;Real-time detection is carried out on the production efficiency of electromechanical product, and it is determined whether it is waste electromechanical product;The quantity of waste electromechanical product is counted as a statistical value;The quantity of waste electromechanical product is predicted as a target comparison value;The difference between the statistical value and the target comparison value is calculated, and whether the digital identification of electromechanical product is completed is determined according to whether the difference is within the set error range.The application determines the quantity of waste electromechanical product by real-time detection of the efficiency of electromechanical product, and then determines the predicted value by using a grey model and compares the predicted value with the actual value, so that accurate evaluation and identification of electromechanical product are realized under the premise of fully considering the production efficiency of electromechanical product.

Description

Technical Field

[0001] This invention relates to the field of electromechanical product inspection technology, and in particular to a digital identification method for electromechanical products. Background Technology

[0002] Electromechanical products refer to various agricultural implements, electrical and electronic production equipment, and household appliances manufactured using mechanical, electrical, and electronic equipment. They generally include mechanical equipment, electrical equipment, transportation vehicles, electronic products, electrical appliances, instruments and meters, metal products, and their parts and components.

[0003] As the use and demand for electromechanical products in daily life gradually increase, they have become indispensable equipment in people's work and life. However, against the backdrop of rapid development in design and manufacturing technology, the technical complexity of various electromechanical products is also constantly increasing, making the maintenance and repair of these products increasingly difficult and arduous, while also placing higher demands on their pass rate. Furthermore, the continuous use of electromechanical products affects their production efficiency, and existing testing and certification methods only focus on surface inspection, neglecting production efficiency and failing to provide accurate evaluation and certification. Summary of the Invention

[0004] This invention provides a digital identification method for electromechanical products. By detecting the efficiency of electromechanical products in real time to determine the quantity of waste electromechanical products, a grey model is used to determine the predicted value and compare the predicted value with the actual value. Under the premise of fully considering the production efficiency of electromechanical products, the method can achieve accurate evaluation and identification of electromechanical products.

[0005] This invention provides a digital identification method for electromechanical products, comprising:

[0006] Images of electromechanical products are acquired, and the state of the electromechanical products is identified using a Faster R-CNN object detection network;

[0007] The production efficiency of the electromechanical products is monitored in real time based on their status, and the results are used to determine whether the electromechanical products are scrap or used.

[0008] The quantity of the aforementioned waste electromechanical products is used as the statistical value;

[0009] The grey (GM1,1) model is used to predict the quantity of waste electromechanical products, and the prediction results are used as the target comparison value.

[0010] Calculate the difference between the statistical value and the target comparison value, and determine whether the difference is within the set error range;

[0011] If the difference is not within the set error range, return to the step of real-time detection of the production efficiency of the electromechanical product based on its status;

[0012] If the difference is within the set error range, the digital identification of the electromechanical product is completed.

[0013] Furthermore, in the step of acquiring an image of an electromechanical product and using a Faster R-CNN object detection network to identify the state of the electromechanical product, the Faster R-CNN object detection network includes a region proposal network and a Fast R-CNN detector. The region proposal network predicts the region where the target is located and shares the information with the Fast R-CNN, thereby obtaining the accurate target location by correcting the target region.

[0014] The region proposal network takes an image of arbitrary size as input and outputs a set of rectangular region proposal boxes with target scores. Its loss function is defined as:

[0015]

[0016] Where i is the index of the sliding window anchor in mini-batch processing, p i The predicted probability of the target object is anchori. This is the label value of the actual bounding box; the value is 0 when an anchor has a negative label and 1 otherwise. i To predict the vector of four parametric coordinates of the bounding box, To predict the parametric coordinate vector of the ground truth bounding box corresponding to the bounding box, L cls For the target classification loss, L reg For the target regression loss, N cls N represents the number of rectangular classification layers. reg λ represents the number of rectangular regression layers, and λ is a coefficient, set to 10.

[0017] The regression loss function of the region proposal network is:

[0018]

[0019] The region proposal network is trained end-to-end using backpropagation and stochastic gradient descent. During training, each mini-batch consists of 256 proposals arbitrarily selected from an image, with a positive to negative label ratio of 1:1.

[0020] The Faster R-CNN object detection network is trained on the basis of a pre-trained ResNet model. The parameters of the layers that are common to the region proposal network are directly copied from the parameters pre-trained on ImageNet, and the parameters of the non-common layers are initialized with a Gaussian distribution with a standard deviation of 0.01.

[0021] Further, the step of performing real-time production efficiency monitoring on the electromechanical products based on their status, and determining whether the electromechanical products are scrap electromechanical products based on the monitoring results, includes:

[0022] When the electromechanical product is in an abnormal state, the abnormal information is recorded and transmitted to the administrator in real time.

[0023] When the electromechanical product is in normal condition, the production efficiency of the electromechanical product is monitored in real time, and it is determined whether the monitoring result exceeds the set value.

[0024] If the test result exceeds the set value, the electromechanical product is determined to be a normal electromechanical product.

[0025] If the test result does not exceed the set value, the electromechanical product is determined to be a used electromechanical product.

[0026] Furthermore, in the step of performing real-time production efficiency detection on the electromechanical product when the product is in normal condition and determining whether the detection result exceeds a set value, the real-time production efficiency detection of the electromechanical product includes:

[0027] Each electromechanical product's data acquisition node is responsible for collecting production data in real time according to its own communication protocol.

[0028] Based on the full life cycle analysis of the electromechanical products, the life cycle stages of the electromechanical products are determined;

[0029] The production efficiency η of the electromechanical product is calculated based on the production data at the current life stage. The calculation formula is as follows:

[0030]

[0031] Wherein, parameter E i Y i and X i N represents the consumption of electrical resources, raw material resources, and natural resources for electromechanical products, respectively. i The total number of electromechanical products manufactured;

[0032] The average production efficiency throughout the entire life cycle of the electromechanical product is used as the real-time detection result of production efficiency.

[0033] Furthermore, in the step of predicting the quantity of waste electromechanical products using a grey (GM1,1) model and using the prediction result as the target comparison value, the grey (GM1,1) model includes:

[0034] Definition 1: Let X (0) =(x (0) (1),x (0) (2),…,x (0) If (n) is a non-negative sequence, then X is called X. (1) For X (0) First-order cumulative generation sequence:

[0035] X (1) =(x (1) (1),x (1) (2),…,x (1) (n))

[0036] in,

[0037] Call x (0) (k)+ax (1) (k) = w is the original form of the (GM1,1) model;

[0038] Z (1) For X (1) The sequence generated by the nearest neighbor mean:

[0039] Z (1) =(z (1) (2),z (1) (3),…z (1) (n))

[0040] in,

[0041] Call x (0) (k)+az (1) (k) = w is the mean form of the (GM1,1) model;

[0042] Definition 2: A first-order linear differential equation is called... The mean of the (GM1,1) model is in the form of x. (0) (k)+ax (1) The whitening equation for (k) = w;

[0043] The parameter series is obtained using the least squares method. The definitions of Y and B are as follows:

[0044]

[0045] Definition 3: In the (GM1,1) model, parameter -a is called the development coefficient, and w is called the grey action quantity.

[0046] Theorem: Let the data sequence X (0) It is a non-negative sequence. As defined in Definition 1, the first-order linear differential equation The solution is:

[0047]

[0048] The above formula is also called the time response function, and its time response sequence is:

[0049]

[0050] Further, we can find the cumulative reduction formula in the above equation:

[0051]

[0052] We can obtain the corresponding X (0) Time-response:

[0053]

[0054] Where k = 1, 2, 3, ..., n.

[0055] Furthermore, the mean absolute percentage error e is used. MAPE The accuracy of the gray (GM1,1) model is verified using the following formula:

[0056]

[0057] in, x 0 (i) and Let represent the actual value and the predicted value, respectively, and let e(i) represent the error between the actual value and the predicted value.

[0058] Furthermore, if the difference is within a set error range, before completing the step of digitally identifying the electromechanical product, the method further includes:

[0059] If the difference is determined to be outside the set error range after a set number of consecutive steps, the data of the electromechanical product for the set number of steps is recorded, and the data is used to correct the gray (GM1,1) model.

[0060] The present invention also provides a digital identification device for electromechanical products, comprising:

[0061] An acquisition module is used to acquire images of electromechanical products and to identify the state of the electromechanical products using a Faster R-CNN object detection network;

[0062] The detection module is used to detect the production efficiency of the electromechanical products in real time according to their status, and to determine whether the electromechanical products are scrap electromechanical products based on the detection results.

[0063] The statistics module is used to count the quantity of the waste electromechanical products as statistical values;

[0064] The prediction module is used to predict the quantity of waste electromechanical products using a grey (GM1,1) model, and uses the prediction results as the target comparison value.

[0065] The calculation module is used to calculate the difference between the statistical value and the target comparison value, and to determine whether the difference is within a set error range;

[0066] The return module is used to return the step of real-time detection of the production efficiency of the electromechanical product based on the status of the electromechanical product when the difference is not within the set error range;

[0067] The termination module is used to complete the digital identification of the electromechanical product when the difference is within a set error range.

[0068] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0069] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0070] The beneficial effects of this invention are as follows:

[0071] This invention first uses a Faster R-CNN object detection network to detect the status of electromechanical products. Only when the status of the electromechanical products is determined to be normal is the production efficiency of the products monitored in real time. Based on the production efficiency, the invention determines whether the electromechanical products are qualified or obsolete. Then, the quantity of obsolete electromechanical products is counted as a statistical value and compared with the predicted quantity of obsolete electromechanical products determined in advance by a grey model. If the difference is within the set error range, the detection and identification of the electromechanical products is determined to be accurate. If the difference is not within the set range, the detection and identification of the electromechanical products is determined to be erroneous, and the process returns to the step of monitoring the production efficiency of the electromechanical products in real time for re-detection and identification until the difference is within the error range, ultimately achieving accurate evaluation and identification of the electromechanical products. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0073] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention.

[0074] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0075] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0076] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0077] like Figure 1 As shown, the present invention provides a digital identification method for electromechanical products, comprising:

[0078] S1. Acquire images of electromechanical products and use Faster R-CNN object detection network to identify the state of the electromechanical products.

[0079] The Faster R-CNN object detection network employs selective search to obtain potential object candidate boxes in an image, and then uses classification and merging methods to obtain the object detection results. The Faster R-CNN object detection network includes a region proposal network (RPN) and a Fast R-CNN detector. The RPN predicts the region where the target is located and shares this information with Fast R-CNN, obtaining the accurate target location by refining the target region.

[0080] The region proposal network is a fully convolutional network that takes an image of arbitrary size as input and outputs a set of rectangular region proposal boxes with target scores. Its loss function is defined as:

[0081]

[0082] Where i is the index of the sliding window anchor in mini-batch processing, p i The predicted probability of the target object is anchori.

[0083] This is the label value of the actual bounding box; the value is 0 when an anchor has a negative label and 1 otherwise. i To predict the vector of four parametric coordinates of the bounding box, To predict the parametric coordinate vector of the ground truth bounding box corresponding to the bounding box, L cls For the target classification loss, L reg For the target regression loss, N clsN represents the number of rectangular classification layers. reg λ represents the number of rectangular regression layers, and λ is a coefficient, set to 10.

[0084] The regression loss function of the region proposal network is:

[0085]

[0086] The region proposal network (RPN) is trained end-to-end using backpropagation and stochastic gradient descent. During training, each mini-batch consists of 256 proposals arbitrarily selected from an image, with a 1:1 ratio of positive to negative labels. In practice, the number of positive samples is often insufficient; in such cases, the number of negative samples should be increased to maintain 256 proposals. Furthermore, some RPN proposal regions may overlap. To reduce redundancy, the network employs non-maximum suppression based on classification scores for filtering. Generally, an overlap rate threshold of 0.7 is set, retaining only candidate boxes with values ​​higher than this.

[0087] The Faster R-CNN object detection network is trained on the basis of a pre-trained ResNet model. The parameters of the layers that are common to the region proposal network are directly copied from the parameters pre-trained on ImageNet, and the parameters of the non-common layers are initialized with a Gaussian distribution with a standard deviation of 0.01.

[0088] S2. Real-time detection of the production efficiency of the electromechanical products based on their status, and determination of whether the electromechanical products are scrap electromechanical products based on the detection results;

[0089] S21. When the electromechanical product is in an abnormal state, the electromechanical product is recorded and the abnormal information is transmitted to the administrator in real time, so that the administrator can quickly find the corresponding electromechanical product for maintenance.

[0090] S22. When the electromechanical product is in normal condition, the production efficiency of the electromechanical product is detected in real time, and it is determined whether the detection result exceeds the set value.

[0091] Real-time monitoring of the production efficiency of the aforementioned electromechanical products includes:

[0092] S221. Each electromechanical product's data acquisition node is responsible for collecting production data of each electromechanical product in real time according to its own communication protocol.

[0093] Data collection rules are predefined in the background, and collection nodes subscribe to message queues via rule messages. Upon initial startup, the collection node requests the collection rule data from the central platform's data acquisition and receiving service module, and then saves the received rule data to a designated directory for the collection module to read. A unified process is adopted to collect various signals from electromechanical product monitoring terminals, divided into four parts: data acquisition, processing, display, and storage.

[0094] The performance of electromechanical products largely depends on the values ​​of capacitors and resistors in the circuit. Choosing an appropriate RC value can effectively improve the efficiency of production data acquisition. As the RC value increases, the signal sampling error decreases, and the accuracy of the sampling results improves. However, an excessively large RC value will significantly delay the acquisition speed. Therefore, all signals below this value will be filtered out. By substituting relevant parameters, the real-time acquisition results of the electromechanical product's production data can be obtained.

[0095] S222. Based on the full life analysis of the electromechanical product, determine the life stage of the electromechanical product;

[0096] In electromechanical product systems, fossil fuels and electrical energy are present throughout the entire product lifecycle. Each lifecycle stage is accompanied by the emission of environmental pollutants such as waste gas and wastewater. During raw material acquisition, raw materials can be obtained from natural resources such as ores, or recycled materials, reusable parts, and remanufactured parts generated during product disposal can be used in the raw material acquisition stage. Considering the entire lifecycle of electromechanical products, it can be divided into the initial stage, growth stage, maturity stage, and decline stage.

[0097] The initial phase mainly involves demand analysis and planning, followed by application, delivery, installation, and commissioning, followed by rigorous acceptance. The mid-life phase, or operation and maintenance phase, can be further divided into the growth and maturity phases, during which the electromechanical product officially enters operational status. Theoretically, the maturity phase represents the highest production efficiency for electromechanical products, where a collaborative operating mode is established between components, and friction between elements gradually decreases. The later stages of the product lifecycle can be divided into the maintenance and scrapping phases. The maintenance phase occurs when multiple anomalies and malfunctions occur while the product is still operational, requiring repeated interruptions to production processes and consuming significant financial and material resources for repairs. During the scrapping phase, the electromechanical product is in a non-operational state.

[0098] Based on the analysis of the life stages of electromechanical products, the characteristics of the electromechanical products working in different life stages can be obtained and set as the quantitative feature vector of product operation: connect the sensor device to the electromechanical product in use, calculate the similarity between the current operating characteristics of the electromechanical product and the set characteristics; compare the calculated similarity result with the similarity threshold value. If the calculated result is higher than the threshold value, it means that the current electromechanical product is in that life stage, and the operating life stage of the electromechanical product can be identified.

[0099] S223. According to the definition of production efficiency, the production efficiency of electromechanical products is the ratio of the effect obtained per unit of time, manpower, raw materials, and capital input during the product's operation; that is, the ratio of output to input. In the definition of production efficiency, input corresponds to the operating resources of electromechanical products, while output corresponds to the function achieved by the production of electromechanical products. The process of utilizing resources is the process of transforming input into output.

[0100] Therefore, the production efficiency η of the electromechanical product is calculated based on the production data at the current life stage, and the calculation formula is as follows:

[0101]

[0102] Wherein, parameter E i Y i and X i N represents the consumption of electrical resources, raw material resources, and natural resources for electromechanical products, respectively. i This refers to the total quantity of electromechanical products manufactured. Because electromechanical products vary in shape, size, and manufacturing complexity, the unit of measurement is used as the unit of measurement in actual calculations.

[0103] S224. The average production efficiency of the electromechanical product throughout its entire life cycle is used as the real-time detection result of the production efficiency.

[0104] The shorter the service life of electromechanical products, the less time is spent on testing, reducing resource occupancy and ensuring successful testing under the same conditions. Since energy efficiency testing is typically conducted every 100 hours, time must be prioritized, and successful energy efficiency testing during operation is more significantly affected. In the real-time monitoring of electromechanical product production efficiency, the product's lifespan status is determined, and production efficiency is calculated based on production data under this status. After collecting data from multiple different stages and averaging the calculated production efficiency, the final real-time production efficiency test result is obtained.

[0105] S23. If the test result exceeds the set value, the electromechanical product is determined to be a normal electromechanical product.

[0106] S24. If the test result does not exceed the set value, then the electromechanical product is determined to be a waste electromechanical product.

[0107] S3. Count the quantity of the waste electromechanical products as statistical values;

[0108] S4. Use the grey (GM1,1) model to predict the quantity of waste electromechanical products and use the prediction results as the target comparison value.

[0109] The gray (GM1,1) model includes:

[0110] Definition 1: Let X (0) =(x (0) (1),x (0) (2),…,x (0) If (n) is a non-negative sequence, then X is called X. (1) For X (0) First-order cumulative generation sequence:

[0111] X (1) =(x (1) (1),x (1) (2),…,x (1) (n))

[0112] in,

[0113] Call x (0) (k)+ax (1) (k) = w is the original form of the (GM1,1) model;

[0114] Z (1) For X (1) The sequence generated by the nearest neighbor mean:

[0115] Z (1) =(z (1) (2),z (1) (3),…z (1) (n))

[0116] in,

[0117] Call x (0) (k)+az (1) (k) = w is the mean form of the (GM1,1) model;

[0118] Definition 2: A first-order linear differential equation is called... The mean of the (GM1,1) model is in the form of x. (0) (k)+az (1) The whitening equation for (k) = w;

[0119] The parameter series is obtained using the least squares method. The definitions of Y and B are as follows:

[0120]

[0121] Definition 3: In the (GM1,1) model, parameter -a is called the development coefficient, and w is called the grey action quantity.

[0122] Theorem: Let the data sequence X (0) It is a non-negative sequence. As defined in Definition 1, the first-order linear differential equation The solution is:

[0123]

[0124] The above formula is also called the time response function, and its time response sequence is:

[0125]

[0126] Further, we can find the cumulative reduction formula in the above equation:

[0127]

[0128] We can obtain the corresponding X (0) Time-response:

[0129]

[0130] Where k = 1, 2, 3, ..., n.

[0131] Using mean absolute percentage error e MAPE The accuracy of the gray (GM1,1) model is verified using the following formula:

[0132]

[0133] in, x 0 (i) and Let represent the actual value and the predicted value, respectively, and let e(i) represent the error between the actual value and the predicted value.

[0134] e MAPE The accuracy class table is as follows:

[0135] <![CDATA[e MAPE Scope Prediction accuracy ≤10％ high 10％～20％ good 20％～50％ feasible ≥50％ Low

[0136] The above values ​​can be adjusted according to actual needs, and no specific limits are set here.

[0137] Example:

[0138] For example, the quantities of scrap electromechanical products from 2012 to 2020 were 172.35, 188.11, 169.68, 208.07, 272.38, 320.91, 450.69, 516.05, and 575.31, respectively (unit: ten thousand).

[0139] The initial data sequence is obtained from the above data:

[0140] X (0) =(172.35,188.11,169.68,208.07,272.38,320.91,450.69,516.05,575.31)

[0141] X (0) Accumulated (1-AGO) generation sequence:

[0142] X (1) =(172.350,360.460,530.140,738.210,1010.590,1331.500,1782.190,2298.240,2873.550)

[0143] Calculate X (1) The sequence generated by the nearest neighbor mean:

[0144] Z (1) =(266.405,445.300,634.175,874.400,1171.045,1556.845,2040.215,2585.895)

[0145] The (GM1,1) model was used to fit the quantity of waste electromechanical products from 2012 to 2020, and the quantity of waste electromechanical products from 2021 to 2023 was predicted. The prediction accuracy of the model was then verified.

[0146] According to Definition 3, the development parameters a and gray action w of the gray model can be obtained: a = -0.19; w = 109.53.

[0147] According to formulas (1) to (4), the GM(1,1) model simulation data of the quantity of waste electromechanical products from 2012 to 2020 is calculated by programming MATLAB software, and the residual and relative error between the simulation data and the actual data are calculated according to formula (5).

[0148] Calculate the average relative error of the simulation data, e MAPE =7.94%, the accuracy level of the simulated data is high.

[0149] The quantity of waste electromechanical products in the next three years is predicted based on the constructed GM(1,1) model, and the prediction accuracy is analyzed by comparing it with the actual quantity of waste electromechanical products in 2021-2023.

[0150] S5. Calculate the difference between the statistical value and the target comparison value, and determine whether the difference is within the set error range;

[0151] S6. If the difference is not within the set error range, it is determined that there is an error in the detection and identification of the electromechanical products. Therefore, the process returns to the step of real-time detection of the production efficiency of the electromechanical products based on their status, and the quantity of waste electromechanical products is re-detected and counted, and then compared again to determine whether it is within the error range. This process is repeated until the difference between the statistical value and the target comparison value is within the error range.

[0152] Furthermore, if the difference is determined to be outside the set error range after a set number of consecutive tests (e.g., five times), the data of the electromechanical product is recorded for five tests, and the gray (GM1,1) model is corrected using this data so that the predicted value of the gray (GM1,1) model is more accurate.

[0153] S7. If the difference is within the set error range, the digital identification of the electromechanical product is completed.

[0154] This invention first uses a Faster R-CNN object detection network to detect the status of electromechanical products. Only when the status of the electromechanical products is determined to be normal is the production efficiency of the products monitored in real time. Based on the production efficiency, the invention determines whether the electromechanical products are qualified or obsolete. Then, the quantity of obsolete electromechanical products is counted as a statistical value and compared with the predicted quantity of obsolete electromechanical products determined in advance by a grey model. If the difference is within the set error range, the detection and identification of the electromechanical products is determined to be accurate. If the difference is not within the set range, the detection and identification of the electromechanical products is determined to be erroneous, and the process returns to the step of monitoring the production efficiency of the electromechanical products in real time for re-detection and identification until the difference is within the error range, ultimately achieving accurate evaluation and identification of the electromechanical products.

[0155] like Figure 2 As shown, the present invention also provides a digital identification device for electromechanical products, comprising:

[0156] Acquisition module 1 is used to acquire images of electromechanical products and use a Faster R-CNN object detection network to identify the state of the electromechanical products;

[0157] The detection module 2 is used to detect the production efficiency of the electromechanical products in real time according to their status, and to determine whether the electromechanical products are scrap electromechanical products based on the detection results.

[0158] Statistical module 3 is used to count the quantity of the waste electromechanical products as statistical values;

[0159] Prediction module 4 is used to predict the quantity of waste electromechanical products using a grey (GM1,1) model, and uses the prediction results as the target comparison value.

[0160] The calculation module 5 is used to calculate the difference between the statistical value and the target comparison value, and to determine whether the difference is within the set error range;

[0161] Return module 6 is used to return the step of real-time detection of production efficiency of the electromechanical product based on the status of the electromechanical product when the difference is not within the set error range;

[0162] End module 7 is used to complete the digital identification of the electromechanical product when the difference is within the set error range.

[0163] In one embodiment, in the acquisition module 1, the Faster R-CNN object detection network includes a region proposal network and a Fast R-CNN detector. The region proposal network predicts the region where the target is located and shares the information with Fast R-CNN. The accurate target location is obtained by correcting the target region.

[0164] The region proposal network takes an image of arbitrary size as input and outputs a set of rectangular region proposal boxes with target scores. Its loss function is defined as:

[0165]

[0166] Where i is the index of the sliding window anchor in mini-batch processing, p i The predicted probability of the target object is anchori. This is the label value of the actual bounding box; the value is 0 when an anchor has a negative label and 1 otherwise. i To predict the vector of four parametric coordinates of the bounding box, To predict the parametric coordinate vector of the ground truth bounding box corresponding to the bounding box, L cls For the target classification loss, L reg For the target regression loss, N cls N represents the number of rectangular classification layers. reg λ represents the number of rectangular regression layers, and λ is a coefficient, set to 10.

[0167] The regression loss function of the region proposal network is:

[0168]

[0169] The region proposal network is trained end-to-end using backpropagation and stochastic gradient descent. During training, each mini-batch consists of 256 proposals arbitrarily selected from an image, with a positive to negative label ratio of 1:1.

[0170] The Faster R-CNN object detection network is trained on the basis of a pre-trained ResNet model. The parameters of the layers that are common to the region proposal network are directly copied from the parameters pre-trained on ImageNet, and the parameters of the non-common layers are initialized with a Gaussian distribution with a standard deviation of 0.01.

[0171] In one embodiment, the detection module 2 step includes:

[0172] The transmission unit is used to record the status of the electromechanical product when it is abnormal, and to transmit the abnormal information to the administrator in real time.

[0173] The detection unit is used to detect the production efficiency of the electromechanical product in real time when the electromechanical product is in normal condition, and to determine whether the detection result exceeds the set value.

[0174] The first determining unit is used to determine that the electromechanical product is a normal electromechanical product when the detection result exceeds a set value;

[0175] The second determining unit is used to determine that the electromechanical product is a waste electromechanical product when the detection result does not exceed the set value.

[0176] In one embodiment, the detection unit performs real-time production efficiency detection on the electromechanical product, including:

[0177] The data acquisition subunit is responsible for collecting production data from each electromechanical product in real time according to its respective communication protocol.

[0178] The analysis subunit is used to determine the life stage of the electromechanical product based on the full life analysis of the electromechanical product.

[0179] The calculation subunit is used to calculate the production efficiency η of the electromechanical product based on the production data at the current life stage. The calculation formula is as follows:

[0180]

[0181] Wherein, parameter E i Y i and X iN represents the consumption of electrical resources, raw material resources, and natural resources for electromechanical products, respectively. i The total number of electromechanical products manufactured;

[0182] The result subunit is used to take the average production efficiency of the electromechanical product throughout its entire life cycle as the real-time detection result of the production efficiency.

[0183] In one embodiment, the gray (GM1,1) model in prediction module 4 includes:

[0184] Definition 1: Let X (0) =(x (0) (1),x (0) (2),…,x (0) If (n) is a non-negative sequence, then X is called X. (1) For X (0) First-order cumulative generation sequence:

[0185] X (1) =(x (1) (1),x (1) (2),…,x (1) (n))

[0186] in,

[0187] Call x (0) (k)+ax (1) (k) = w is the original form of the (GM1,1) model;

[0188] Z (1) For X (1) The sequence generated by the nearest neighbor mean:

[0189] Z (1) =(z (1) (2),z (1) (3),…z (1) (n))

[0190] in,

[0191] Call x (0) (k)+az (1) (k) = w is the mean form of the (GM1,1) model;

[0192] Definition 2: A first-order linear differential equation is called... The mean of the (GM1,1) model is in the form of x. (0) (k)+az (1) The whitening equation for (k) = w;

[0193] The parameter series is obtained using the least squares method. The definitions of Y and B are as follows:

[0194]

[0195] Definition 3: In the (GM1,1) model, parameter -a is called the development coefficient, and w is called the grey action quantity.

[0196] Theorem: Let the data sequence X (0) It is a non-negative sequence. As defined in Definition 1, the first-order linear differential equation The solution is:

[0197]

[0198] The above formula is also called the time response function, and its time response sequence is:

[0199]

[0200] Further, we can find the cumulative reduction formula in the above equation:

[0201]

[0202] We can obtain the corresponding X (0) Time-response:

[0203]

[0204] Where k = 1, 2, 3, ..., n.

[0205] In one embodiment, the mean absolute percentage error e is used. MAPE The accuracy of the gray (GM1,1) model is verified using the following formula:

[0206]

[0207] in, x 0 (i) and Let represent the actual value and the predicted value, respectively, and let e(i) represent the error between the actual value and the predicted value.

[0208] In one embodiment, it also includes:

[0209] The correction module is used to record the data of the electromechanical product for a set number of consecutive times when the difference is determined to be outside the set error range, and to use it to correct the gray (GM1,1) model.

[0210] Each of the above modules, units, and sub-units is used to perform the respective steps in the above-mentioned digital identification method for electromechanical products. The specific implementation methods are as described in the above-mentioned method embodiments, and will not be repeated here.

[0211] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all data required for the process of digital identification of electromechanical products. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the digital identification method for electromechanical products.

[0212] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0213] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for digital identification of electromechanical products.

[0214] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media provided in this application and in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0215] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0216] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for digital identification of electromechanical products, characterized in that, include: Images of electromechanical products are acquired, and the state of the products is identified using a Faster R-CNN object detection network. The Faster R-CNN object detection network includes a region proposal network and a Fast R-CNN detector. The region proposal network predicts the region where the target is located and shares this information with the Fast R-CNN, obtaining the accurate target location by correcting the target region. The region proposal network takes an image of arbitrary size as input and outputs a set of rectangular region proposal boxes with target scores. Its loss function is defined as: in, This refers to the index of the sliding window anchor in mini-batch processing. for anchor The predicted probability of the target object, This is the label value of the actual bounding box. The value is 0 when an anchor has a negative label and 1 otherwise. To predict the vector of four parametric coordinates of the bounding box, To predict the parametric coordinate vector of the ground truth bounding box corresponding to the bounding box, Classification loss for the target, For the target regression loss, The number of rectangular classification layers, This represents the number of rectangular regression layers. The coefficient is set to 10; the regression loss function of the region proposal network is: The region proposal network is trained end-to-end using backpropagation and stochastic gradient descent. During training, each mini-batch consists of 256 proposals arbitrarily selected from an image, with a 1:1 ratio of positive to negative labels. The Faster R-CNN object detection network is trained on a pre-trained ResNet model. The parameters of the layers shared with the region proposal network are directly copied from the parameters pre-trained on ImageNet, while the parameters of the non-shared layers are initialized using a Gaussian distribution with a standard deviation of 0.

01. The production efficiency of the electromechanical products is monitored in real time based on their status, and the results are used to determine whether the electromechanical products are scrap or used. The quantity of the aforementioned waste electromechanical products is used as the statistical value; The grey (GM1,1) model is used to predict the quantity of waste electromechanical products, and the prediction results are used as the target comparison value. Calculate the difference between the statistical value and the target comparison value, and determine whether the difference is within the set error range; If the difference is not within the set error range, return to the step of real-time detection of the production efficiency of the electromechanical product based on its status; If the difference is within the set error range, the digital identification of the electromechanical product is completed.

2. The digital identification method for electromechanical products according to claim 1, characterized in that, The step of performing real-time production efficiency monitoring on the electromechanical products based on their status, and determining whether the electromechanical products are scrap electromechanical products based on the monitoring results, includes: When the electromechanical product is in an abnormal state, the abnormal information is recorded and transmitted to the administrator in real time. When the electromechanical product is in normal condition, the production efficiency of the electromechanical product is monitored in real time, and it is determined whether the monitoring result exceeds the set value. If the test result exceeds the set value, the electromechanical product is determined to be a normal electromechanical product. If the test result does not exceed the set value, the electromechanical product is determined to be a used electromechanical product.

3. The digital identification method for electromechanical products according to claim 1, characterized in that, If the difference is within a set error range, the process before completing the digital authentication of the electromechanical product further includes: If the difference is determined to be outside the set error range after a set number of consecutive steps, the data of the electromechanical product for the set number of steps is recorded, and the data is used to correct the gray (GM1,1) model.

4. A digital identification device for electromechanical products, characterized in that, include: An acquisition module is used to acquire images of electromechanical products and identify the state of the products using a Faster R-CNN object detection network. The Faster R-CNN object detection network includes a region proposal network and a Fast R-CNN detector. The region proposal network predicts the region where the target is located and shares this information with the Fast R-CNN, obtaining the accurate target location by correcting the target region. The region proposal network takes an image of arbitrary size as input and outputs a set of rectangular region proposal boxes with target scores. Its loss function is defined as: in, This refers to the index of the sliding window anchor in mini-batch processing. for anchor The predicted probability of the target object, This is the label value of the actual bounding box. The value is 0 when an anchor has a negative label and 1 otherwise. To predict the vector of four parametric coordinates of the bounding box, To predict the parametric coordinate vector of the ground truth bounding box corresponding to the bounding box, Classification loss for the target, For the target regression loss, The number of rectangular classification layers, This represents the number of rectangular regression layers. The coefficient is set to 10; the regression loss function of the region proposal network is: The region proposal network is trained end-to-end using backpropagation and stochastic gradient descent. During training, each mini-batch consists of 256 proposals arbitrarily selected from an image, with a 1:1 ratio of positive to negative labels. The Faster R-CNN object detection network is trained on a pre-trained ResNet model. The parameters of the layers shared with the region proposal network are directly copied from the parameters pre-trained on ImageNet, while the parameters of the non-shared layers are initialized using a Gaussian distribution with a standard deviation of 0.

01. The detection module is used to detect the production efficiency of the electromechanical products in real time according to their status, and to determine whether the electromechanical products are scrap electromechanical products based on the detection results. The statistics module is used to count the quantity of the waste electromechanical products as statistical values; The prediction module is used to predict the quantity of waste electromechanical products using a grey (GM1,1) model, and uses the prediction results as the target comparison value. The calculation module is used to calculate the difference between the statistical value and the target comparison value, and to determine whether the difference is within a set error range; The return module is used to return the step of real-time detection of the production efficiency of the electromechanical product based on the status of the electromechanical product when the difference is not within the set error range; The termination module is used to complete the digital identification of the electromechanical product when the difference is within a set error range.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 3.

Images