Non-woven fabric production defect autonomous early warning method based on multi-feature composite identification

Abstract

The invention discloses a non-woven fabric production defect autonomous early warning method based on multi-feature composite identification, and relates to the technical field of non-woven fabric production, and the method comprises the steps: obtaining at least one production step; obtaining at least one flaw feature of the non-woven fabric; classifying the flaw features into non-visual flaw features and visual flaw features; associating the production steps with the non-visual defect features and the visual defect features; forming a non-visual identification scheme for the non-visual defect features; forming a first important weight for the non-visual defect features, and forming a second important weight for the visual defect features; forming a normal critical range of the non-woven fabric; the flaw severity coefficient of the non-woven fabric is calculated; and carrying out early warning on the target non-visual defect features and the target visual defect features. A non-visual identification scheme is formed by classifying flaw features and non-visual flaw features, so that flaw omission can be avoided, and a generated result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of nonwoven fabric production, and in particular to an autonomous early warning method for nonwoven fabric production defects based on multi-feature composite recognition. Background Art

[0002] Non-woven fabrics are fiber materials that do not need to be woven into cloth during the production process. They are usually made of synthetic fibers such as polyester, nylon, and polypropylene, or natural fibers. In the process of manufacturing non-woven fabrics, some defects will occur on the production line, which will affect the appearance quality and performance of the product. These defects are complex and may be visible defects or invisible characteristics such as poor water absorption, low strength, and easy disconnection.

[0003] The existing technology lacks consideration for the above-mentioned complex defect composition, and its early warning for non-woven fabric production is insufficiently complete. It is easy to miss early warnings for defects, and may also give wrong early warnings, affecting the efficiency of normal production. Summary of the invention

[0004] In order to solve the above technical problems, a method for autonomous early warning of non-woven fabric production defects based on multi-feature composite recognition is provided. This technical solution solves the problem that the existing technology proposed in the above background technology lacks consideration for the above-mentioned complex defect composition, and its early warning for non-woven fabric production is not complete enough, and it is easy to miss early warnings for defects, and may also give wrong early warnings, affecting the efficiency of normal production.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] An autonomous early warning method for nonwoven production defects based on multi-feature composite recognition, comprising:

[0007] Decomposing the nonwoven fabric production process to obtain at least one production step;

[0008] Based on historical inspection data of the non-woven fabric, obtaining at least one defect feature of the non-woven fabric;

[0009] Classifying defect features into non-visual defect features and visible defect features;

[0010] Associating production steps with non-visual defect characteristics and visible defect characteristics;

[0011] Forming non-visual recognition schemes for non-visual defect features;

[0012] Visual defect features are identified using vision, and non-visual defect features are identified using non-visual recognition solutions;

[0013] The first important weight is formed for the non-visual defect features, and the second important weight is formed for the visible defect features;

[0014] Based on historical data, a normal critical range for nonwoven fabrics is formed;

[0015] In a production step of the nonwoven fabric, first actual data of non-visual defect characteristics are obtained, and second actual data of visible defect characteristics are obtained;

[0016] Calculate the defect severity coefficient of the non-woven fabric. When the defect severity coefficient does not fall within the normal critical range, the production step is regarded as an abnormal production step.

[0017] Acquire non-visual defect features and visible defect features associated with the abnormal production step as target non-visual defect features and target visible defect features;

[0018] Based on historical data, early warning is provided for target non-visual defect features and target visible defect features.

[0019] Preferably, the nonwoven fabric production process is decomposed to obtain at least one production step, which comprises the following steps:

[0020] Evenly divide the nonwoven fabric production process in chronological order to obtain at least one partial production process;

[0021] Determine whether the production processes used in the adjacent partial production processes are the same, if so, merge the adjacent partial production processes, if not, do nothing;

[0022] When all adjacent partial production processes are judged, the combined partial production processes are used as production steps.

[0023] Preferably, the step of obtaining at least one defect feature of the nonwoven fabric comprises the following steps:

[0024] Based on historical detection data of the non-woven fabric, obtaining at least one detection index of the non-woven fabric;

[0025] Using one of the at least one detection indicator as a target detection indicator;

[0026] Acquire at least one target nonwoven fabric, satisfying that the values ​​of the target detection indicators of the at least one target nonwoven fabric are different, and the values ​​of the remaining detection indicators of the at least one target nonwoven fabric are all within the normal value range of the corresponding detection indicators;

[0027] Counting the unqualified ratio of at least one target non-woven fabric, when the unqualified ratio is greater than a preset ratio, the target detection index is used as a defect feature, otherwise, no processing is performed;

[0028] When the target detection index traverses at least one detection index, at least one defect feature is obtained.

[0029] Preferably, the classifying the defect features into non-visual defect features and visible defect features comprises the following steps:

[0030] Obtaining at least one qualified non-woven fabric and at least one unqualified non-woven fabric;

[0031] Performing image recognition on at least one qualified non-woven fabric to obtain at least one first feature, which is summarized into a first feature set, wherein the image recognition is performed by means of visible light;

[0032] Performing image recognition on at least one unqualified non-woven fabric to obtain at least one second feature, which is summarized into a second feature set;

[0033] Obtaining the intersection of the first feature set and the second feature set as the third feature set;

[0034] Performing a difference set operation on the second feature set and the third feature set to obtain a fourth feature set;

[0035] Elements in the fourth feature set are used as visible defect features, and defect features other than the visible defect features in the at least one defect feature are used as non-visual defect features.

[0036] Preferably, associating the production steps with the non-visual defect characteristics and the visible defect characteristics comprises the following steps:

[0037] The parameters of the production step are adjusted, and the changed non-visual defect characteristics and visible defect characteristics are associated with the production step.

[0038] Preferably, the forming of a non-visual identification scheme for non-visual defect features comprises the following steps:

[0039] A first sample nonwoven fabric and a second sample nonwoven fabric having different values ​​of non-visual defect features are obtained, and other parameter conditions of the first sample nonwoven fabric and the second sample nonwoven fabric are consistent;

[0040] Based on the big data, at least one ray is obtained, where the ray is non-visible light;

[0041] Screening out a ray that will not damage the nonwoven fabric from at least one ray as a target ray;

[0042] Using target rays to develop and image the first sample nonwoven fabric to obtain a first sample image, and using target rays to develop and image the second sample nonwoven fabric to obtain a second sample image;

[0043] When the first sample image is inconsistent with the second sample image, the process of using the target ray for recognition is used as a non-visual recognition scheme.

[0044] Preferably, forming the first important weight for the non-visual defect feature and forming the second important weight for the visible defect feature comprises the following steps:

[0045] averaging the values ​​of the non-visual defect feature of at least one qualified non-woven fabric to obtain a first average value;

[0046] averaging the values ​​of the visual defect feature of at least one qualified nonwoven fabric to obtain a second average value;

[0047] Acquire at least one first non-woven fabric, satisfying that the difference between the value of the non-visual defect characteristic of the first non-woven fabric and the first average value is greater than a preset difference;

[0048] Counting the unqualified proportion of at least one first non-woven fabric as a first important weight;

[0049] Acquire at least one second non-woven fabric, where the difference between the value of the visible defect feature of the second non-woven fabric and the second mean value is greater than a preset difference;

[0050] The unqualified ratio of at least one second nonwoven fabric is counted as the second important weight.

[0051] Preferably, the forming of the normal critical range of the nonwoven fabric based on historical data comprises the following steps:

[0052] Based on the historical data, obtaining historical data of non-visual defect characteristics and historical data of visual defect characteristics of at least one qualified non-woven fabric;

[0053] Use the first defect formula to calculate the defect severity coefficient of the qualified non-woven fabric;

[0054] The range formed by the maximum value and the minimum value of the defect severity coefficient of at least one qualified non-woven fabric is used as the normal critical range of the non-woven fabric;

[0055] The first defect formula is as follows:

[0056]

[0057] Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, and a i is the historical data of non-visual defect features, b i is the first important weight of non-visual defect features, c j is the historical data of visual defect characteristics, d j is the second most important weight of the visual defect feature.

[0058] Preferably, the calculation of the defect severity coefficient of the nonwoven fabric comprises the following steps:

[0059] Use the second defect formula to calculate the defect severity coefficient of the nonwoven fabric;

[0060] The second defect formula is as follows:

[0061]

[0062] Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, e is the total number of non-visual defect features, and i is the first actual data of non-visual defect features, b i is the first important weight of non-visual defect features, f j is the second actual data of the visible defect feature, d j is the second most important weight of the visual defect feature.

[0063] Preferably, the step of providing an early warning for target non-visual defect features and target visible defect features based on historical data comprises the following steps:

[0064] Calculate the ratio of the difference between the first actual data of the target non-visual defect feature and the first mean value of the target non-visual defect feature to the first mean value of the target non-visual defect feature as the first deviation;

[0065] Calculating the ratio of the difference between the second actual data of the target visual defect feature and the second mean value of the target visual defect feature to the second mean value of the target visual defect feature as the second deviation;

[0066] An early warning is issued for a target non-visual defect feature whose first deviation is greater than a preset deviation and a target visible defect feature whose second deviation is greater than a preset deviation.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] By classifying defect features, associating production steps with non-visual defect features and visible defect features, forming a non-visual recognition scheme for non-visual defect features, and forming a first important weight for non-visual defect features and a second important weight for visual defect features, it is possible to use a non-contact method to perform defect detection on non-woven fabric production, which can detect both visible and invisible features, thereby avoiding missing defects, and combining the results of multiple non-visual defect features and visual defect features based on the first important weight and the second important weight, so that the generated results are more accurate, thereby avoiding early warning errors and thus avoiding production stagnation, and when abnormalities occur, accurate early warnings of abnormal features can be provided, thereby enabling targeted adjustments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1It is a schematic diagram of the process of the autonomous early warning method for non-woven fabric production defects based on multi-feature composite recognition of the present invention;

[0070] Figure 2 The nonwoven fabric production process of the present invention is decomposed to obtain a flow chart of at least one production step;

[0071] Figure 3 A schematic diagram of a process for obtaining at least one defect feature of a nonwoven fabric according to the present invention;

[0072] Figure 4 A schematic diagram of a flow chart of classifying defect features into non-visual defect features and visible defect features according to the present invention;

[0073] Figure 5 A schematic diagram of a process for forming a non-visual identification solution for non-visual defect features of the present invention;

[0074] Figure 6 A schematic diagram of a process of forming a first important weight for a non-visual defect feature and a second important weight for a visible defect feature according to the present invention;

[0075] Figure 7 A schematic diagram of a process for forming a normal critical range of a nonwoven fabric based on historical data of the present invention;

[0076] Figure 8 The figure is a flow chart of the present invention for early warning of target non-visual defect features and target visible defect features based on historical data. DETAILED DESCRIPTION

[0077] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0078] Reference Figure 1 As shown, a non-woven fabric production defect autonomous early warning method based on multi-feature composite recognition includes:

[0079] Decomposing the nonwoven fabric production process to obtain at least one production step;

[0080] Based on historical inspection data of the non-woven fabric, obtaining at least one defect feature of the non-woven fabric;

[0081] Classifying defect features into non-visual defect features and visible defect features;

[0082] Associating production steps with non-visual defect characteristics and visible defect characteristics;

[0083] Forming non-visual recognition schemes for non-visual defect features;

[0084] Visual defect features are identified using vision, and non-visual defect features are identified using non-visual recognition solutions;

[0085] The first important weight is formed for the non-visual defect features, and the second important weight is formed for the visible defect features;

[0086] Based on historical data, a normal critical range for nonwoven fabrics is formed;

[0087] In a production step of the nonwoven fabric, first actual data of non-visual defect characteristics are obtained, and second actual data of visible defect characteristics are obtained;

[0088] Calculate the defect severity coefficient of the non-woven fabric. When the defect severity coefficient does not fall within the normal critical range, the production step is regarded as an abnormal production step.

[0089] Acquire non-visual defect features and visible defect features associated with the abnormal production step as target non-visual defect features and target visible defect features;

[0090] Based on historical data, early warning is provided for target non-visual defect features and target visible defect features.

[0091] In this scheme, the main purpose is to identify and warn possible defects in the production process of non-woven fabrics through non-contact detection. Since the defect characteristics may be invisible under normal conditions, it is necessary to distinguish them and form a non-visual recognition scheme for invisible defect characteristics. On the other hand, different non-visual defect characteristics and visible defect characteristics have different effects on non-woven fabrics. Therefore, in order to integrate the results of multiple characteristics, it is necessary to weight the non-visual defect characteristics and the visual defect characteristics. This scheme calculates the proportion to more accurately form the first important weight for non-visual defect characteristics and the second important weight for visual defect characteristics. In addition, the usual warning only makes an alarm but does not point out the specific characteristics of the problem. Therefore, it is necessary to check one by one, which will increase the time of production stagnation. In this scheme, the non-visual defect characteristics and visual defect characteristics that contribute more to the abnormality can be pointed out, so that the scope of investigation is narrowed, thereby reducing the time of production stagnation caused by investigation.

[0092] Reference Figure 2 As shown, the nonwoven fabric production process is decomposed to obtain at least one production step including the following steps:

[0093] Evenly divide the nonwoven fabric production process in chronological order to obtain at least one partial production process;

[0094] Determine whether the production processes used in the adjacent partial production processes are the same, if so, merge the adjacent partial production processes, if not, do nothing;

[0095] When all adjacent partial production processes are judged, the combined partial production processes are used as production steps.

[0096] The production steps only include a single production process. Therefore, when early warning is performed in each production step, the influence of too many processes involved can be avoided, so that the detection for early warning is more targeted. When the local production process is small enough, the local production process only includes one production process, because the non-woven fabric production process is carried out according to one production process;

[0097] When merging, the adjacent local production processes in at least one local production process are merged according to the above rules, and the adjacent local production processes in at least one local production process after the merger are merged again according to the above rules, and the cycle is repeated until no adjacent local production processes are merged.

[0098] Reference Figure 3 As shown, obtaining at least one defect feature of the nonwoven fabric includes the following steps:

[0099] Based on historical detection data of the non-woven fabric, obtaining at least one detection index of the non-woven fabric;

[0100] Using one of the at least one detection indicator as a target detection indicator;

[0101] Acquire at least one target nonwoven fabric, satisfying that the values ​​of the target detection indicators of the at least one target nonwoven fabric are different, and the values ​​of the remaining detection indicators of the at least one target nonwoven fabric are all within the normal value range of the corresponding detection indicators;

[0102] Counting the unqualified ratio of at least one target non-woven fabric, when the unqualified ratio is greater than a preset ratio, the target detection index is used as a defect feature, otherwise, no processing is performed;

[0103] When the target detection index traverses at least one detection index, at least one defect feature is obtained.

[0104] There are many detection indicators for non-woven fabrics, but not all of them have an impact on the quality of non-woven fabrics. Therefore, in order to reduce the workload of early warning, it is necessary to screen the detection indicators of non-woven fabrics to obtain defect characteristics, and the early warning is only performed on the defect characteristics. As a result, the amount of identified data will be reduced, which can improve the efficiency of early warning.

[0105] Reference Figure 4 As shown, classifying defect features into non-visual defect features and visible defect features includes the following steps:

[0106] Obtaining at least one qualified non-woven fabric and at least one unqualified non-woven fabric;

[0107] Performing image recognition on at least one qualified non-woven fabric to obtain at least one first feature, which is summarized into a first feature set, wherein the image recognition is performed by means of visible light;

[0108] Performing image recognition on at least one unqualified non-woven fabric to obtain at least one second feature, which is summarized into a second feature set;

[0109] Obtaining the intersection of the first feature set and the second feature set as the third feature set;

[0110] Performing a difference set operation on the second feature set and the third feature set to obtain a fourth feature set;

[0111] Elements in the fourth feature set are used as visible defect features, and defect features other than the visible defect features in the at least one defect feature are used as non-visual defect features.

[0112] The principle is that visual defect features must be different under visible light conditions. Therefore, when the features that are different from those in qualified non-woven fabrics in unqualified non-woven fabrics are visual defect features, that is, elements in the fourth feature set, and defect features are either non-visual defect features or visible defect features, therefore, defect features other than visual defect features are non-visual defect features.

[0113] Associating production steps with non-visual defect characteristics and visible defect characteristics includes the following steps:

[0114] The parameters of the production step are adjusted, and the changed non-visual defect characteristics and visible defect characteristics are associated with the production step.

[0115] The purpose of the association is: when abnormal production steps are obtained, only the non-visual defect features and visible defect features associated with the abnormal production steps can be screened for early warning, without the need to screen all non-visual defect features and visual defect features, which can further reduce the workload of early warning.

[0116] Reference Figure 5 As shown, forming a non-visual recognition scheme for non-visual defect features includes the following steps:

[0117] A first sample nonwoven fabric and a second sample nonwoven fabric having different values ​​of non-visual defect features are obtained, and other parameter conditions of the first sample nonwoven fabric and the second sample nonwoven fabric are consistent;

[0118] Based on the big data, at least one ray is obtained, where the ray is non-visible light;

[0119] Screening out a ray that will not damage the nonwoven fabric from at least one ray as a target ray;

[0120] Using target rays to develop and image the first sample nonwoven fabric to obtain a first sample image, and using target rays to develop and image the second sample nonwoven fabric to obtain a second sample image;

[0121] When the first sample image is inconsistent with the second sample image, the process of using the target ray for recognition is used as a non-visual recognition scheme.

[0122] Non-visual defect features will not appear under visible light conditions, but will definitely appear under certain light conditions. Therefore, the exhaustive and variable control methods are used to screen the rays, thereby obtaining target rays that can identify non-visual defect features.

[0123] Reference Figure 6 As shown, forming the first important weight for the non-visual defect feature and forming the second important weight for the visual defect feature includes the following steps:

[0124] averaging the values ​​of the non-visual defect feature of at least one qualified non-woven fabric to obtain a first average value;

[0125] averaging the values ​​of the visual defect feature of at least one qualified nonwoven fabric to obtain a second average value;

[0126] Acquire at least one first non-woven fabric, satisfying that the difference between the value of the non-visual defect characteristic of the first non-woven fabric and the first average value is greater than a preset difference;

[0127] Counting the unqualified proportion of at least one first non-woven fabric as a first important weight;

[0128] Acquire at least one second non-woven fabric, where the difference between the value of the visible defect feature of the second non-woven fabric and the second mean value is greater than a preset difference;

[0129] The unqualified ratio of at least one second nonwoven fabric is counted as the second important weight.

[0130] The weight formation process of non-visual defect features and visible defect features is similar. Therefore, only the weighting of non-visual defect features is explained. There are many non-visual defect features in non-woven fabrics, not just one non-visual defect feature. Therefore, it is difficult to directly determine the weight of a single non-visual defect feature based on its unqualified situation. In order to determine the weight of a single non-visual defect feature, when the value of the non-visual defect feature is relatively abnormal, the unqualified ratio of at least one first non-woven fabric is obtained as the first important weight. Since the weight of each of the remaining non-visual defect features is obtained by this method, under this calculation method, the interference factors other than a single non-visual defect feature can be regarded as consistent. Therefore, the only difference is the degree of abnormality of a single non-visual defect feature. Therefore, the larger the first important weight, the greater the influence of a single non-visual defect feature. This is in line with common sense. Therefore, the first important weight can be used as the weight of the non-visual defect feature.

[0131] Reference Figure 7 As shown, based on historical data, the normal critical range for forming nonwoven fabrics includes the following steps:

[0132] Based on the historical data, obtaining historical data of non-visual defect characteristics and historical data of visual defect characteristics of at least one qualified non-woven fabric;

[0133] Use the first defect formula to calculate the defect severity coefficient of the qualified non-woven fabric;

[0134] The range formed by the maximum value and the minimum value of the defect severity coefficient of at least one qualified non-woven fabric is used as the normal critical range of the non-woven fabric;

[0135] The first defect formula is as follows:

[0136]

[0137] Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, and a i is the historical data of non-visual defect features, b i is the first important weight of non-visual defect features, c j is the historical data of visual defect characteristics, d j is the second most important weight of the visual defect feature.

[0138] The normal critical range is used to determine whether the defect severity coefficient is abnormal or not.

[0139] Calculating the defect severity factor for nonwoven fabrics involves the following steps:

[0140] Use the second defect formula to calculate the defect severity coefficient of the nonwoven fabric;

[0141] The second defect formula is as follows:

[0142]

[0143] Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, e is the total number of non-visual defect features, and i is the first actual data of non-visual defect features, b i is the first important weight of non-visual defect features, f j is the second actual data of the visible defect feature, d j is the second most important weight of the visual defect feature.

[0144] The degree of influence of different non-visual defect features and visible defect features, therefore, their data must be weighted when integrated, otherwise the accuracy of the results will be limited.

[0145] Reference Figure 8 As shown, based on historical data, early warning of target non-visual defect features and target visible defect features includes the following steps:

[0146] Calculate the ratio of the difference between the first actual data of the target non-visual defect feature and the first mean value of the target non-visual defect feature to the first mean value of the target non-visual defect feature as the first deviation;

[0147] Calculating the ratio of the difference between the second actual data of the target visual defect feature and the second mean value of the target visual defect feature to the second mean value of the target visual defect feature as the second deviation;

[0148] An early warning is issued for a target non-visual defect feature whose first deviation is greater than a preset deviation and a target visible defect feature whose second deviation is greater than a preset deviation.

[0149] Among the abnormal influences, the larger the deviation of the target non-visual defect features and target visible defect features from the mean is, the greater the abnormal contribution they produce. Therefore, if the early warning is performed on these features, it can not only make adjustments according to the inspection results during the inspection and solve the existing anomalies, but also reduce the number of features to be inspected, thereby improving the efficiency of detection.

[0150] Furthermore, the present solution also proposes a storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned non-woven fabric production defect autonomous early warning method based on multi-feature composite recognition is executed.

[0151] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state drive (SSD).

[0152] To sum up, the advantages of the present invention are: by classifying defect features, associating production steps with non-visual defect features and visible defect features, forming a non-visual recognition scheme for non-visual defect features, and forming a first important weight for non-visual defect features and a second important weight for visual defect features, it is possible to use a non-contact method to perform defect detection on non-woven fabric production, and it is possible to detect both visible features and invisible features at the same time, thereby avoiding missing defects, and according to the first important weight and the second important weight, the results of multiple non-visual defect features and visual defect features are integrated, so that the generated results are more accurate, thereby avoiding early warning errors and thus avoiding production stagnation.

[0153] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. An autonomous early warning method for non-woven fabric production defects based on multi-feature composite recognition, characterized in that: include: Decomposing the nonwoven fabric production process to obtain at least one production step; Based on historical inspection data of the non-woven fabric, obtaining at least one defect feature of the non-woven fabric; Classifying defect features into non-visual defect features and visible defect features; Associating production steps with non-visual defect characteristics and visible defect characteristics; Forming non-visual recognition schemes for non-visual defect features; Visual defect features are identified using vision, and non-visual defect features are identified using non-visual recognition solutions; The first important weight is formed for the non-visual defect features, and the second important weight is formed for the visible defect features; Based on historical data, a normal critical range for nonwoven fabrics is formed; In a production step of the nonwoven fabric, first actual data of non-visual defect characteristics are obtained, and second actual data of visible defect characteristics are obtained; Calculate the defect severity coefficient of the non-woven fabric. When the defect severity coefficient does not fall within the normal critical range, the production step is regarded as an abnormal production step. Acquire non-visual defect features and visible defect features associated with the abnormal production step as target non-visual defect features and target visible defect features; Based on historical data, early warning is provided for target non-visual defect features and target visible defect features.

2. The method for autonomous early warning of nonwoven production defects based on multi-feature composite recognition according to claim 1 is characterized in that: Decomposing the nonwoven fabric production process to obtain at least one production step includes the following steps: Evenly divide the nonwoven fabric production process in chronological order to obtain at least one partial production process; Determine whether the production processes used in the adjacent partial production processes are the same, if so, merge the adjacent partial production processes, if not, do nothing; When all adjacent partial production processes are judged, the combined partial production processes are used as production steps.

3. The method for autonomous early warning of nonwoven fabric production defects based on multi-feature composite recognition according to claim 2 is characterized in that: The obtaining of at least one defect feature of the nonwoven fabric comprises the following steps: Based on historical detection data of the non-woven fabric, obtaining at least one detection index of the non-woven fabric; Using one of the at least one detection indicator as a target detection indicator; Acquire at least one target nonwoven fabric, satisfying that the values ​​of the target detection indicators of the at least one target nonwoven fabric are different, and the values ​​of the remaining detection indicators of the at least one target nonwoven fabric are all within the normal value range of the corresponding detection indicators; Counting the unqualified ratio of at least one target non-woven fabric, when the unqualified ratio is greater than a preset ratio, the target detection index is used as a defect feature, otherwise, no processing is performed; When the target detection index traverses at least one detection index, at least one defect feature is obtained.

4. The method for autonomous early warning of nonwoven fabric production defects based on multi-feature composite recognition according to claim 3 is characterized in that: The classification of defect features into non-visual defect features and visible defect features comprises the following steps: Obtaining at least one qualified non-woven fabric and at least one unqualified non-woven fabric; Performing image recognition on at least one qualified non-woven fabric to obtain at least one first feature, which is summarized into a first feature set, wherein the image recognition is performed by means of visible light; Performing image recognition on at least one unqualified non-woven fabric to obtain at least one second feature, which is summarized into a second feature set; Obtaining the intersection of the first feature set and the second feature set as the third feature set; Performing a difference set operation on the second feature set and the third feature set to obtain a fourth feature set; Elements in the fourth feature set are used as visible defect features, and defect features other than the visible defect features in the at least one defect feature are used as non-visual defect features.

5. The method for autonomous early warning of nonwoven fabric production defects based on multi-feature composite recognition according to claim 4 is characterized in that: The step of associating the production steps with the non-visual defect characteristics and the visible defect characteristics comprises the following steps: The parameters of the production step are adjusted, and the changed non-visual defect characteristics and visible defect characteristics are associated with the production step.

6. The method for autonomous early warning of nonwoven fabric production defects based on multi-feature composite recognition according to claim 5, characterized in that: The non-visual identification scheme for non-visual defect features comprises the following steps: A first sample nonwoven fabric and a second sample nonwoven fabric having different values ​​of non-visual defect features are obtained, and other parameter conditions of the first sample nonwoven fabric and the second sample nonwoven fabric are consistent; Based on the big data, at least one ray is obtained, where the ray is non-visible light; Screening out a ray that will not damage the nonwoven fabric from at least one ray as a target ray; Using target rays to develop and image the first sample nonwoven fabric to obtain a first sample image, and using target rays to develop and image the second sample nonwoven fabric to obtain a second sample image; When the first sample image is inconsistent with the second sample image, the process of using the target ray for recognition is used as a non-visual recognition scheme.

7. The method for autonomous early warning of nonwoven production defects based on multi-feature composite recognition according to claim 6 is characterized in that: The forming of the first important weight for the non-visual defect feature and the forming of the second important weight for the visual defect feature comprises the following steps: averaging the values ​​of the non-visual defect feature of at least one qualified non-woven fabric to obtain a first average value; averaging the values ​​of the visual defect feature of at least one qualified nonwoven fabric to obtain a second average value; Acquire at least one first non-woven fabric, satisfying that the difference between the value of the non-visual defect characteristic of the first non-woven fabric and the first average value is greater than a preset difference; Counting the unqualified proportion of at least one first non-woven fabric as a first important weight; Acquire at least one second non-woven fabric, where the difference between the value of the visible defect feature of the second non-woven fabric and the second mean value is greater than a preset difference; The unqualified ratio of at least one second nonwoven fabric is counted as the second important weight.

8. The method for autonomous early warning of nonwoven production defects based on multi-feature composite recognition according to claim 7, characterized in that: The formation of a normal critical range for nonwoven fabrics based on historical data includes the following steps: Based on the historical data, obtaining historical data of non-visual defect characteristics and historical data of visual defect characteristics of at least one qualified non-woven fabric; Use the first defect formula to calculate the defect severity coefficient of the qualified non-woven fabric; The range formed by the maximum value and the minimum value of the defect severity coefficient of at least one qualified non-woven fabric is used as the normal critical range of the non-woven fabric; The first defect formula is as follows: Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, and a i is the historical data of non-visual defect features, b i is the first important weight of non-visual defect features, c j is the historical data of visual defect characteristics, d j is the second most important weight of the visual defect feature.

9. The method for autonomous early warning of nonwoven fabric production defects based on multi-feature composite recognition according to claim 8, characterized in that: The calculation of the defect severity coefficient of the nonwoven fabric comprises the following steps: Use the second defect formula to calculate the defect severity coefficient of the nonwoven fabric; The second defect formula is as follows: Among them, A is the defect severity coefficient of qualified non-woven fabrics, i and j are subscripts, n is the total number of non-visual defect features, m is the total number of visible defect features, e is the total number of non-visual defect features, and i is the first actual data of non-visual defect features, b i is the first important weight of non-visual defect features, f j is the second actual data of the visible defect feature, d j is the second most important weight of the visual defect feature.

10. The method for autonomous early warning of nonwoven production defects based on multi-feature composite recognition according to claim 9, characterized in that: The method of providing an early warning for target non-visual defect features and target visible defect features based on historical data includes the following steps: Calculate the ratio of the difference between the first actual data of the target non-visual defect feature and the first mean value of the target non-visual defect feature to the first mean value of the target non-visual defect feature as the first deviation; Calculating the ratio of the difference between the second actual data of the target visual defect feature and the second mean value of the target visual defect feature to the second mean value of the target visual defect feature as the second deviation; An early warning is issued for a target non-visual defect feature whose first deviation is greater than a preset deviation and a target visible defect feature whose second deviation is greater than a preset deviation.

Images