Defective product recall method and early warning method

Abstract

The application discloses a kind of defective product recall method and early warning method, by adding sensor to equipment, and the data collected is persisted.By mining sensor data features and product defect features, the correlation between the two is analyzed, a product defect trend curve is constructed, and the product defect change process is analyzed.This method can deeply mine historical processing data, can track products of different defect types, has a wide range of application scenarios and high reference value.By constructing a defect product failure trend curve and setting a threshold, the production period of defective products is determined, which is more accurate than traditional defective product recall methods (product batch recall), avoiding under-recall and over-recall of products, reducing losses for enterprises;A variety of methods are used to construct a defect product failure trend curve, each method can be used alone or in combination (such as weighted average), with higher accuracy.

Description

Technical Field

[0001] This invention relates to the field of discrete manufacturing technology, and in particular to a defective product recall method and early warning method that can ensure accurate and rapid recall of current defective products, while preventing the recurrence of similar defects in the future. Background Technology

[0002] In discrete manufacturing enterprises, a comprehensive defective product recall mechanism is essential. A sound recall mechanism can effectively address the safety hazards caused by product defects. When industrial equipment (such as machine tools and industrial robots) malfunctions during processing, the manufactured products may be in the early stages of defective products (minor defects) or have hidden defects (not covered by current detection mechanisms or assessment methods). These products are highly likely to end up in the hands of consumers. Only when the defects become severe or hidden defects are exposed do companies begin to formulate product recall plans. The time from the generation of defective products to the implementation of defective product recalls can take months or years, making it difficult to accurately assess the health status of products in the hands of consumers. Therefore, effectively assessing the health status of historically manufactured products is crucial for accurately implementing defective product recalls, avoiding under-recalls and over-recalls, and reducing unnecessary losses for both enterprises and users.

[0003] Patent application number 201510132302.9 describes a "product recall warning system and method based on QR codes." This method establishes a unique QR code for each product, recording relevant product information such as the batch number to which the product belongs. When a defect occurs in a product of the same type and requires recall, that product falls within the scope of the defective product recall. Main drawbacks: 1. Classifying products by batch number results in a crude method for identifying defective products. When defective products exist across multiple batches or are only a portion of a batch, the product defect recall method of this patent is ineffective. 2. This method uses user feedback information as a feature for product defect recall, which cannot meet the needs of defect recalls for high-precision machined parts in fields such as machine tools and automobile manufacturing. 3. For periodic processing equipment failures, not all processed parts are defective; there is no in-depth analysis of whether a product is defective and the degree of defect.

[0004] Patent application number 202011422718.1 describes a "business processing method, apparatus, and device based on product recall." This method discloses a business processing method, apparatus, and device based on product recall, comprising the following steps: receiving a product recall business; the product recall business includes information about products to be recalled; querying at least one product recall station corresponding to the information about products to be recalled; the product recall station is used to transmit and / or store the products to be recalled; sending a product recall instruction to a recall node device corresponding to the product recall station; the product recall instruction is used to instruct the product recall station to recall the products to be recalled.

[0005] Disadvantages: 1. This method primarily addresses the business processing of product recalls, mainly improving the automation level of the recall process, simplifying the recall steps, and shortening the product recall time. However, this patent does not accurately determine whether a product is defective, nor does it assess the degree of product defect. 2. This product recall mechanism does not create a knowledge base for defective products, does not save the experience that led to the defective product as an asset, and lacks an alarm mechanism when similar defective products reappear.

[0006] It is evident that most companies currently employ a simple and crude method for recalling defective products, with the main process as follows: 1. Identify defective products using user feedback; 2. Trace the batch to which the product belongs and related product batches; 3. Implement a recall for the relevant product batches. This method is inefficient, lacks accuracy, and has a long cycle from problem discovery to recall implementation.

[0007] In summary, the main drawbacks of the existing technology are as follows:

[0008] 1. The current method of recalling defective products is mainly based on the batch of defective products. Recalls are implemented on the batches of defective products that have been found. The batches of products to which no defects have been found cannot be said to be without defects. They may be overlooked due to insufficient sampling or the degree of defect being not obvious. Without effective assessment of product defects, it is easy to lead to over-recall and under-recall of products.

[0009] 2. Without classifying the degree of defect in defective products and determining the priority of product recall, users are more likely to suffer losses due to untimely recall of severely defective products.

[0010] 3. There was no further analysis or exploration of the defective product characteristics to form an experience knowledge base, and no early warning mechanism for similar defective products to occur again. Summary of the Invention

[0011] In view of the above problems, the present invention provides a defective product recall method and early warning method for overcoming or at least partially solving the above problems.

[0012] This invention provides the following solution:

[0013] A method for recalling defective products includes:

[0014] Sensor data from the target sensor is collected according to the target time granularity and stored in chronological order; the target sensor includes several sensors installed on several components of the product processing equipment, and the several sensors are used to sense the operating status of the several components in a one-to-one correspondence.

[0015] Extract several sensor fault characteristics contained in the sensor data;

[0016] Determine the time point at which the defective product sample was found; and determine the first time period for producing the defective product based on the preset time range;

[0017] Select several first target fault features from several sensor fault features within the first time period that can effectively distinguish the product defect status, and sort the several first target fault features according to the importance of the fault features.

[0018] For each of the first target fault features, trend analysis and segmentation are performed to obtain the segmentation of several first target fault features near the first time period and the importance of the fault features, so as to determine the upper and lower bounds of the second time period in which the defective product is located.

[0019] Determine several second target fault features included in the second time period, and sort the several second target fault features according to the importance of the fault features;

[0020] Several second target fault features are smoothed and mapped to the 0-100 range to construct a product health score curve;

[0021] The defective product levels are classified according to the product health score curve.

[0022] The categories and priorities of product recalls are determined based on the results of the defective product classification.

[0023] Preferably, the target sensor includes any one or more of a vibration sensor, a flow sensor, and a temperature sensor.

[0024] Preferably, the target sensor includes a vibration sensor, and the sensor fault characteristics include one or more of the following: time domain characteristics, frequency domain characteristics, signal decomposition characteristics, spectral amplitude characteristics, and envelope characteristics after Hilbert transform.

[0025] Preferably, the preset time range is determined based on the actual situation of the equipment processing category.

[0026] Preferably, the first target fault feature is a feature that can effectively distinguish the product defect state from several sensor fault features obtained in the first time period by using variance analysis or feature selection based on classification models.

[0027] Preferably, the method for determining the upper and lower bounds of the second time period includes:

[0028] The sliding window weighted counting method is used to determine the time window r, calculate whether each fault feature under the sliding window has slice segmentation, and statistically analyze the results.

[0029] The formula for determining whether a given window represents the upper or lower bound of the time period containing defective products is as follows:

[0030] Vj = sum(wi * Fi j), where i = 1, 2, ..., m, j = 1, 2, ..., S

[0031] Based on the previous iteration results t1 and t2, select the window containing the maximum Vj near t1 and t2, and update t1 and t2 using the window's midpoint time t1` and t2`, i.e., t1 = t1` and t2 = t2`.

[0032] Preferably, the product health score curve is a score curve generated individually based on a certain fault feature or a weighted fusion of multiple fault features.

[0033] Preferably, the product health score curve is constructed using an exponential mapping method.

[0034] Preferably, the exponential mapping method includes the following formula:

[0035] score=a*e b*x

[0036] Wherein, a and b are parameters, and the values ​​of parameters a and b are determined based on the fault characteristic values ​​of existing defective products.

[0037] A defective product early warning method includes:

[0038] The above-mentioned defective product recall methods are saved to form a product health score model, and product defect category labels are marked.

[0039] All fault characteristics used in the existing model are calculated using sensor data collected during the current processing of the product.

[0040] Call the saved models to calculate the product health score curve and obtain the health score of the currently processed product;

[0041] An alarm mechanism is generated based on the current health score of the processed products.

[0042] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] This application provides a defective product recall and early warning method, which involves installing sensors on equipment and persistently storing the collected data. By mining the characteristics of sensor data and product defect characteristics, the correlation between the two is analyzed to construct a product defect trend curve and analyze the process of product defect changes. This method can deeply mine historical processing data, track products with different defect types, and has a wide range of applications and high reference value.

[0044] By constructing a defective product failure trend curve and setting a preset threshold, the production period of defective products can be determined. This method is more accurate than the traditional defective product recall method (product batch recall), avoiding under-recall and over-recall of products and reducing losses for enterprises. Multiple methods are used to construct the defective product failure trend curve. Each method can be used alone or multiple methods can be combined (such as weighted average), resulting in higher accuracy.

[0045] With only a small sample of defective products, it is possible to uncover the degree of product defects at different times, identify products with minor defects in the early stages, and classify the degree of product defects according to the processing time of the equipment (e.g., minor defects, moderate defects, and severe defects). Based on the degree of defect of the defective products, the priority of product recall can be determined, reducing the losses caused to users by products with severe defects.

[0046] Furthermore, in the preferred embodiment, the defective product early warning method provided in this application can provide an early warning when the defective product reappears in the future by retaining the fault characteristics or fault trend curve of each type of defective product, thereby reducing the production of moderately and severely defective products and having greater economic value for enterprises.

[0047] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0049] Figure 1 This is a flowchart of a defective product recall method provided in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of fault feature extraction provided in an embodiment of the present invention;

[0051] Figure 3 This is an overall block diagram of defective product recall provided in an embodiment of the present invention;

[0052] Figure 4 This is a flowchart of a defective product early warning method provided in an embodiment of the present invention;

[0053] Figure 5 This is a defective product early warning block diagram provided in an embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0055] See Figure 1 This invention provides a method for recalling defective products, such as... Figure 1 As shown, the method may include:

[0056] S101: Collect sensor data of the target sensor according to the target time granularity and store it in time sequence; the target sensor includes several sensors installed on several components of the product processing equipment, and the several sensors are used to sense the operating status of the several components in a one-to-one correspondence; specifically, the target sensor includes any one or more of vibration sensors, flow sensors, and temperature sensors.

[0057] In practical applications, depending on the type of equipment, fault detection devices (including but not limited to vibration sensors, current and voltage sensors, temperature sensors, flow sensors, etc.) are installed, and sensor data is uploaded to the data center according to different target time granularities (such as 1 minute, 10 minutes, etc.).

[0058] S102: Extract several sensor fault features contained in the sensor data; the target sensor includes a vibration sensor, and the sensor fault features include one or more of the following: time domain features, frequency domain features, signal decomposition features, spectral amplitude features, and envelope features after Hilbert transform.

[0059] S103: Determine the time point where the defective product sample has been found; and determine the first time period for producing the defective product according to the preset time range; specifically, the preset time range is determined according to the actual situation of the equipment processing category.

[0060] S104: Select several first target fault features from several sensor fault features within the first time period that can effectively distinguish the product defect state, and sort the several first target fault features according to the importance of the fault features; specifically, the first target fault features are features that can effectively distinguish the product defect state from several sensor fault features obtained within the first time period through variance analysis or feature selection based on classification models.

[0061] Based on the discovered defective products, sensor data features (such as the mean value of vibration sensors, the average temperature of temperature sensors, etc.) containing the time period (defective products) are extracted. Using the data of this time period as negative samples, a semi-supervised model is constructed. By continuously extracting and selecting fault features, the sample labels of the time period are adjusted. The maximum difference between positive and negative sample features is used as the evaluation index to select fault features.

[0062] S105: Perform trend analysis and segmentation on each of the first target fault features to obtain the segmentation status of several first target fault features near the first time period and the importance of the fault features, so as to determine the upper and lower bounds of the second time period in which the defective product is located; specifically, the method for determining the upper and lower bounds of the second time period includes:

[0063] The sliding window weighted counting method is used to determine the time window r, calculate whether each fault feature under the sliding window has slice segmentation, and statistically analyze the results.

[0064] The formula for determining whether a given window represents the upper or lower bound of the time period containing defective products is as follows:

[0065] Vj = sum(wi * Fi j), where i = 1, 2, ..., m, j = 1, 2, ..., S

[0066] Based on the previous iteration results t1 and t2, select the window containing the maximum Vj near t1 and t2, and update t1 and t2 using the window's midpoint time t1` and t2`, i.e., t1 = t1` and t2 = t2`.

[0067] S106: Determine a number of second target fault features included in the second time period, and sort the number of second target fault features according to the importance of the fault features;

[0068] S107: After smoothing several second target fault features, map them to the 0-100 range to construct a product health score curve; specifically, the product health score curve is a score curve generated individually based on a certain fault feature or a weighted fusion of multiple fault features.

[0069] In one implementation, embodiments of this application may provide a product health score curve constructed using an exponential mapping method.

[0070] Specifically, the exponential mapping method includes the following formula:

[0071] score=a*e b*x

[0072] Wherein, a and b are parameters, and the values ​​of parameters a and b are determined based on the fault characteristic values ​​of existing defective products.

[0073] S108: Classify the defective product level according to the product health score curve;

[0074] S109: Determine the category and priority of product recall based on the defective product classification results. Specifically, construct a product health status scoring curve (out of 100) using the failure characteristic trend curve. Classify the product defect severity according to the scoring curve (e.g., greater than 80 points - qualified product; less than 40 points - severe defect; 40-60 points - moderate defect; 60-80 points - minor defect, etc.). Determine the product recall priority according to the severity of the product defect.

[0075] The core technology of defective product recall lies in the accurate identification of defective products, the assessment of defect severity, and the construction of an early warning mechanism based on defective product characteristics. This ensures the accurate and rapid recall of current defective products while preventing similar defects from recurring, thus reducing unnecessary losses for enterprises and users. The defective product recall method provided in this application abandons the batch-based recall method. By installing sensors (such as vibration sensors and temperature sensors) on equipment and persistently storing the collected data, when a defect occurs in a processed product, it may not be detected due to the technology or usage scenario at the time. However, after a period of time (such as 1 month, 6 months, or 1 year), the defective product is gradually discovered. At this point, by mining the characteristics of sensor data and product defect characteristics, the correlation between the two is analyzed to construct a product defect trend curve and analyze the process of product defect change. This method can deeply mine historical processing data, track products with different defect types, and has a wide range of applications and high reference value.

[0076] The following section uses the acquisition of vibration sensor data as an example to provide a detailed description of the defective product recall method provided in this application embodiment.

[0077] Depending on the specific processing equipment, appropriate sensors (such as vibration sensors, flow sensors, and temperature sensors) are installed. For CNC machine tools (such as grinding machines), vibration sensors and temperature sensors, among others, are installed on the spindle box. The collected sensor data is uploaded to the database and persistently stored at different time granularities (such as 1 minute, 10 minutes, etc.). This provides effective data support for the construction of models for subsequent offline feature extraction, product health curve analysis, defective product recall, and defective product early warning.

[0078] The defective product recall process is as follows:

[0079] Step 1: Sensor Data Feature Extraction

[0080] like Figure 2 As shown, the main feature of sensor data is time series, the difference being that the sampling frequency of each sensor is different. Taking vibration data as an example, features are extracted.

[0081] Data preprocessing. Extract sensor vibration time-series data from the data center and perform preprocessing (such as outlier handling, missing value handling, filtering, detrending, etc.).

[0082] Fault feature extraction. This involves extracting features from the vibration signal in the time domain, frequency domain, and signal decomposition domains. Features include, but are not limited to, the following:

[0083] 1. Time-domain characteristics of vibration data, including but not limited to mean, root mean square, standard deviation, kurtosis, margin factor, etc.;

[0084] 2. Vibration data spectrum amplitude characteristics, including but not limited to peak amplitude and mean amplitude;

[0085] 3. Frequency domain characteristics of vibration data, including but not limited to center of gravity frequency, main frequency band energy characteristics, and spectrum dispersion;

[0086] 4. Vibration signal decomposition features, such as wavelet packet decomposition and empirical mode decomposition, are used to further extract features from the decomposed signal, such as extracting energy features of each frequency band after wavelet packet decomposition.

[0087] 5. Feature extraction after vibration signal transformation, such as envelope features after Hilbert transform.

[0088] Step 2: Marking the characteristics of the time period of the defective product

[0089] Based on samples of defective products, the production time period of these defective products is determined using methods such as tracking and location (e.g., batch positioning). The initial time point t0 of the defective product is determined, and products produced within a preset time range T (e.g., 1 hour, 12 hours, 1 day, 7 days) are tentatively classified as defective. That is, the time range for defective products is (t0-T / 2, t0+T / 2). Products produced outside this time range are considered non-defective. The value of this time range T is selected based on actual conditions such as the type of equipment being processed.

[0090] Step 3: Sensor Fault Feature Selection

[0091] Based on the sensor features extracted in Step 1 and the labeled sample data, fault features that can effectively distinguish product defect states are initially selected. Methods for fault feature selection include, but are not limited to, analysis of variance (ANOVA) and feature selection based on classification models. Simultaneously, based on the variance or model results, the importance of each fault feature can be obtained, and the fault features are sorted from largest to smallest according to their importance, with corresponding importance values ​​w1, w2, w3…wn (w1>w2>w3>…>wn). The main purpose of this step is to recall fault features with fault discriminative power as much as possible.

[0092] Step 4: Defective product labeling optimization and fault feature selection

[0093] For each fault feature in step 3, perform trend analysis and segmentation (segmentation methods include, but are not limited to, thresholding, clustering, etc.). Considering the segmentation of all selected fault features within the time interval (t0-T / 2, t0+T / 2) and the importance of the fault features, determine the upper and lower bounds (t1, t2) of the time period where the defective product is located. Optimize the labeling of the time period where the defective product is located, and recalculate the fault feature weights. After this step is iterated K times (K is generally 3-4), finally M fault features are obtained, with their importance values ​​w1, w2, w3…wm (w1>w2>w3>…>wm).

[0094] In the above, t1 and t2 can be determined using the sliding window weighted counting method. The time window r is determined, and it is calculated whether each fault feature under the sliding window has a slice segment. If it does, it is recorded as 1; otherwise, it is recorded as 0. The statistical results are shown in Table 1.

[0095] Table 1 Statistical Results of Fault Feature Slices

[0096]

[0097]

[0098] In Table 1, Fij (i = 1, 2…m, j = 1, 2…S) takes the value 0 or 1. The formula for calculating whether a certain window is the upper or lower bound of the time period in which the defective product is located is:

[0099] Vj = sum(wi * Fi j), where i = 1, 2, ..., m, j = 1, 2, ..., S

[0100] Based on the previous iteration results t1 and t2, select the window containing the maximum Vj near t1 and t2, and update t1 and t2 using the window's midpoint times t1` and t2`. That is, t1 = t1` and t2 = t2`.

[0101] Step 5: Feature Smoothing

[0102] The selected fault features in step 4 are smoothed. The smoothing methods include, but are not limited to, sliding window mean smoothing, sliding window mean smoothing, Gaussian smoothing, piecewise fitting smoothing, etc. One or more of these smoothing methods can be selected and fused together.

[0103] Step 6: Construct a product health score curve

[0104] The smoothed feature values ​​are mapped to the range of 0 to 100 to construct a product health score curve. An exponential mapping method can be used to construct the score curve, such as score = a * eb * x, where a and b are parameters, determined based on the fault feature values ​​of existing defective products.

[0105] The product health score curve can be generated individually based on a single fault feature, or it can be a weighted fusion of multiple fault features.

[0106] Step 7: Defective Product Classification

[0107] Based on the product health score curve obtained in step 6, the defective product levels are as follows:

[0108] Table 2 Classification of Defective Product Levels

[0109] Product rating Defective Product Grade 0～40 Severe defect (D) 40～60 Moderate defect (C) 60～80 Minor defect (B) 80～100 Qualified Product (A)

[0110] Overall framework diagram of defective product recall as follows Figure 3 As shown:

[0111] Step 8: Determine the products that need to be recalled and their priority.

[0112] Based on the product defect severity level identified in step 7, determine the product recall category and priority. After determining the degree of product defect, prioritize recalling category D products, followed by category C and then category B products. Alternatively, depending on the actual situation, consider recalling products with a score of 70-80. This method further reduces the impact of severely defective products on users' health or interests due to imperfections in the product recall process or untimely recalls.

[0113] In summary, the defective product recall method provided in this application involves installing sensors on equipment and persistently storing the collected data. By mining the characteristics of sensor data and product defect characteristics, the correlation between the two is analyzed to construct a product defect trend curve and analyze the process of product defect changes. This method can deeply mine historical processing data, track products with different defect types, and has a wide range of applications and high reference value.

[0114] By constructing a defective product failure trend curve and setting a preset threshold, the production period of defective products can be determined. This method is more accurate than the traditional defective product recall method (product batch recall), avoiding under-recall and over-recall of products and reducing losses for enterprises. Multiple methods are used to construct the defective product failure trend curve. Each method can be used alone or multiple methods can be combined (such as weighted average), resulting in higher accuracy.

[0115] With only a small sample of defective products, it is possible to uncover the degree of product defects at different times, identify products with minor defects in the early stages, and classify the degree of product defects according to the processing time of the equipment (e.g., minor defects, moderate defects, and severe defects). Based on the degree of defect of the defective products, the priority of product recall can be determined, reducing the losses caused to users by products with severe defects.

[0116] See Figure 4 This invention provides a defective product early warning method, such as... Figure 4 As shown, the method may include:

[0117] S201: Save the above-mentioned defective product recall methods to form a product health score model, and label the product defect category.

[0118] S202: Calculate all fault characteristics used in the existing model by using the sensor data collected during the current processing of the product;

[0119] S203: Call the saved models, calculate the product health score curve, and obtain the health score of the currently processed product;

[0120] S204: An alarm mechanism is generated based on the current health score of the processed products.

[0121] The defective product early warning method provided in this application persistently stores the defective product category and the corresponding fault feature-scoring model, denoted as product defect category 1-scoring model 1, ..., product defect category N-scoring model N. In the actual production process, the defective products that have occurred are monitored in real time. If a similar defect occurs again, an alarm can be triggered in the early stage of the defective product occurrence.

[0122] The product defect early warning scheme is as follows:

[0123] Step 1: Save the health score model

[0124] During the defective product recall process, the product health score model is saved and the product defect category is labeled.

[0125] Step 2: Real-time calculation of online sensor features

[0126] The collected sensor data is used to calculate all the fault characteristics used in the existing model in real time.

[0127] Step 3: Product Health Score Prediction

[0128] Call up the saved models, calculate the product health score curve, and obtain the health score of the currently processed product.

[0129] Step 4: Determine the product's health status and trigger alarms if necessary.

[0130] Based on the product rating data obtained in step 3, if the rating is below 80 points, a minor alarm will be triggered automatically. The alarm rules are shown in Table 3.

[0131] Table 3 Classification of Early Warning Levels for Defective Products

[0132] Product rating Call the police 0～40 Severe alarm 40～60 Moderate alarm 60～80 Minor alarm 80～100 No police report

[0133] Defective Product Early Warning Diagram as follows Figure 5 As shown:

[0134] In summary, the defective product early warning method provided in this application, by retaining the fault characteristics or fault trend curves of each type of defective product, can provide early warning when the equipment produces the same defective product again in the future, thereby reducing the production of moderately and severely defective products and thus having greater economic value for enterprises.

[0135] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0136] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0137] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0138] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for recalling defective products, characterized in that, include: Sensor data from the target sensor is collected according to the target time granularity and stored in chronological order; the target sensor includes several sensors installed on several components of the product processing equipment, and the several sensors are used to sense the operating status of the several components in a one-to-one correspondence. Extract several sensor fault characteristics contained in the sensor data; Determine the time point at which the defective product sample was found; and determine the first time period for producing the defective product based on the preset time range; Select several first target fault features from several sensor fault features within the first time period that can effectively distinguish the product defect status, and sort the several first target fault features according to the importance of the fault features. For each of the first target fault features, trend analysis and segmentation are performed to obtain the segmentation of several first target fault features near the first time period and the importance of the fault features, so as to determine the upper and lower bounds of the second time period in which the defective product is located. Determine several second target fault features included in the second time period, and sort the several second target fault features according to the importance of the fault features; Several second target fault features are smoothed and mapped to the 0-100 range to construct a product health score curve; The defective product levels are classified according to the product health score curve. The categories and priorities of product recalls are determined based on the results of the defective product classification.

2. The defective product recall method according to claim 1, characterized in that, The target sensor includes any one or more of the following: vibration sensor, flow sensor, and temperature sensor.

3. The defective product recall method according to claim 2, characterized in that, The target sensor includes a vibration sensor, and the sensor fault characteristics include one or more of the following: time domain characteristics, frequency domain characteristics, signal decomposition characteristics, spectral amplitude characteristics, and envelope characteristics after Hilbert transform.

4. The defective product recall method according to claim 1, characterized in that, The preset time range is determined based on the actual situation of the equipment processing category.

5. The defective product recall method according to claim 1, characterized in that, The first target fault feature is a feature that can effectively distinguish the product defect state from several sensor fault features obtained in the first time period by using variance analysis or feature selection based on classification models.

6. The defective product recall method according to claim 1, characterized in that, The methods for determining the upper and lower bounds of the second time period include: The sliding window weighted counting method is used to determine the time window r, calculate whether each fault feature under the sliding window has slice segmentation, and statistically analyze the results. The formula for determining whether a given window represents the upper or lower bound of the time period containing defective products is as follows: Vj = sum(wi*Fij), where i = 1, 2, ..., m, j = 1, 2, ..., S Based on the previous iteration results t1 and t2, select the window containing the maximum Vj near t1 and t2, and update t1 and t2 using the window's midpoint time t1` and t2`, i.e., t1 = t1` and t2 = t2`.

7. The defective product recall method according to claim 1, characterized in that, The product health score curve is a score curve generated individually based on a certain fault feature or a weighted fusion of multiple fault features.

8. The defective product recall method according to claim 7, characterized in that, The product health score curve is constructed using the exponential mapping method.

9. The defective product recall method according to claim 8, characterized in that, The exponential mapping method includes the following formula: score=a*e b*x Wherein, a and b are parameters, and the values ​​of parameters a and b are determined based on the fault characteristic values ​​of existing defective products.

10. A method for early warning of defective products, characterized in that, include: The defective product recall method described in any one of claims 1 to 9 is saved to form a product health score model, and the product defect category is labeled. All fault characteristics used in the existing model are calculated using sensor data collected during the current processing of the product. Call the saved models to calculate the product health score curve and obtain the health score of the currently processed product; An alarm mechanism is generated based on the current health score of the processed products.

Images