Multi-modal data fusion all-cotton spunlace machine fault identification method and system

Abstract

The invention discloses a multi-modal data fusion all-cotton spunlace machine fault identification method and system, and relates to the technical field of data processing. The method comprises the following steps: collecting a water flow sequence, a motor characteristic sequence and a fiber web image sequence of a plurality of spunlace assemblies of the all-cotton spunlace machine; inputting the fiber web image sequence into a streak identification network group, obtaining a streak defect rate sequence through defect identification and statistical analysis, and further calculating a streak implicit failure rate reflecting a defect change trend; calculating a spunlace implicit anomaly rate based on minimum deviation analysis of the water flow and the motor feature sequence; and converting the spunlace implicit abnormal rate and the streak implicit failure rate into a first equipment failure rate and a second equipment failure rate respectively, and obtaining a comprehensive equipment failure rate as an identification result through weighted fusion. According to the invention, early-stage accurate identification and cross verification of the fault of the all-cotton spunlace machine are realized, the limitation of single signal source analysis is overcome, and the accuracy and timeliness of fault diagnosis are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for fault identification of cotton spunlace machines using multimodal data fusion. Background Technology

[0002] Hydroentangled nonwoven fabric production equipment is a key piece of equipment in the textile industry. This equipment typically arranges several hydroentanglement units along the conveyor belt and fabric width, using high-pressure water jets to entangle the fibers and form a stable fabric structure. In actual operation, if a hydroentanglement unit experiences abnormal water flow or motor instability, it will result in insufficient fiber entanglement and a loose structure in the corresponding area. When this area enters the operating range of subsequent normal hydroentanglement units, the high-pressure water jets can easily penetrate the loose areas, causing some fibers to be scattered or piled to the sides, ultimately forming streaks on the finished product surface, severely affecting product quality.

[0003] Currently, fault identification for all-cotton hydrospunlace machines mainly relies on single-parameter threshold judgment or post-production inspection. The former only triggers an alarm when a parameter exceeds a set threshold, making it difficult to detect gradual parameter degradation and early subtle anomalies; the latter only detects defects after they have formed, exhibiting significant lag and resulting in a large number of defective products and resource losses. Existing methods fail to achieve real-time correlation and comprehensive analysis between equipment operating status and product quality characteristics, leading to insufficient accuracy in fault identification and delayed early warnings, thus failing to meet the higher requirements of intelligent manufacturing for predictive maintenance and precise quality control. Summary of the Invention

[0004] This invention addresses the technical problems of low fault identification accuracy, delayed early warning, and failure to comprehensively diagnose equipment operating status and product quality in real time in existing technologies. It provides a multimodal data fusion-based fault identification method and system for cotton spunlace machines.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for fault identification of a cotton spunlace machine based on multimodal data fusion, including: Multimodal data of multiple hydrospun parts during the operation of the all-cotton hydrospun machine were collected to obtain multiple water flow sequences and multiple motor characteristic sequences, and the fiber web image sequence of the all-cotton hydrospun nonwoven fabric was also collected. Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and a streak defect rate sequence is output. The streak hidden failure rate is then obtained through processing. Based on the multiple water flow sequences and multiple motor characteristic sequences, the water jet hidden anomaly rate is calculated; The equipment failure rate is obtained by verifying the implied anomaly rate and the implied failure rate of the spunlace and streak, and is used as the identification result.

[0006] Secondly, this invention provides a multimodal data fusion-based fault identification system for all-cotton spunlace machines, comprising: The data acquisition module is used to collect multimodal data of multiple hydroentanglement components during the operation of the all-cotton hydroentanglement machine, obtain multiple water flow sequences and multiple motor characteristic sequences, and acquire fiber web image sequences of all-cotton hydroentangled nonwoven fabric. The streak analysis module is used to input each fiber web image in the fiber web image sequence into the streak recognition network group, output the streak defect rate sequence, and process it to obtain the streak hidden failure rate. The hydroentanglement status analysis module is used to calculate the hydroentanglement hidden anomaly rate based on the multiple water flow sequences and multiple motor characteristic sequences; The fault identification module is used to verify the equipment failure rate based on the implied anomaly rate of the spunlace and the implied failure rate of the streak, and to obtain the equipment failure rate as the identification result.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention first constructs a dual-perception dimension of equipment status and product quality by collecting multimodal operating data and fiber web image sequences from the spunlace machine unit, providing a more comprehensive data foundation for fault identification. Secondly, it utilizes a streak recognition network group to intelligently analyze the fiber web images, accurately extracting the streak-related implicit fault rate from product appearance features, achieving a quantitative characterization of quality defects and an indirect mapping of equipment health status. Thirdly, it calculates the spunlace-related implicit anomaly rate based on the water flow rate and motor feature sequences of the multi-spunlace components, directly capturing abnormal signs at the equipment operation level, forming an active fault detection capability. Finally, by cross-validating and fusing the spunlace-related implicit anomaly rate and the streak-related implicit fault rate, it achieves the correlation diagnosis between abnormal equipment operating parameters and product quality defects, effectively overcoming the limitations of single signal source analysis, improving the accuracy, timeliness, and reliability of fault identification, and providing technical support for predictive maintenance of cotton spunlace machines. Attached Figure Description

[0008] Figure 1 A flowchart illustrating the multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines provided by this invention; Figure 2 This is a schematic diagram of the multimodal data fusion fault identification system for all-cotton hydroentanglement machines provided by the present invention.

[0009] In the attached diagram, the components represented by each number are as follows: Data acquisition module 11, streak analysis module 12, hydroentanglement status analysis module 13, fault identification module 14. Detailed Implementation

[0010] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0012] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0013] Example 1, as Figure 1 As shown, this embodiment of the invention provides a multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines, including: S10: Collect multimodal data of multiple hydroentanglement components during the operation of the all-cotton hydroentanglement machine, obtain multiple water flow sequences and multiple motor characteristic sequences, and collect fiber web image sequences of all-cotton hydroentangled nonwoven fabric. Multimodal data of multiple hydrospunlace components during the operation of the all-cotton hydrospunlace machine were collected to obtain multiple water flow sequences and multiple motor characteristic sequences. Additionally, a sequence of fiber web images of the all-cotton hydrospunlace nonwoven fabric was acquired, including: Multimodal data of multiple hydrospunlace components were collected during the operation of the all-cotton hydrospunlace machine. The multimodal data included water flow rate and motor characteristics. By integrating multimodal data from multiple hydroentanglement components according to time sequence, multiple water flow sequences and multiple motor characteristic sequences are obtained; Images of multiple web regions along the forward direction during the production of all-cotton spunlace nonwoven fabric were collected to obtain a web image sequence.

[0014] First, multimodal data is collected. During the continuous operation of the all-cotton hydroentangling machine, data is collected from each hydroentangling component distributed across the machine. The collected data includes water flow rate data characterizing the working state of the high-pressure hydroentangling needles and motor characteristic data reflecting the operating status of the hydroentangling unit drive system. Water flow rate refers to the volume of liquid passing through the nozzle of the hydroentangling component per unit time, directly reflecting the energy input intensity of the hydroentangling process. Motor characteristics refer to the physical and electrical characteristics exhibited by the motor driving the hydroentangling component during operation, including various operating parameters such as current, voltage, and vibration. Specifically, real-time monitoring and acquisition of this multimodal data are achieved through flow meters and motor status sensors installed on each hydroentangling component.

[0015] Secondly, the multimodal data collected from multiple spunlace components were integrated and arranged chronologically. For each spunlace component, its water flow data was organized according to the order of its acquisition time, forming a water flow sequence reflecting the time-varying pattern of water flow for that component. Similarly, the motor characteristic data of each spunlace component were organized chronologically, forming a motor characteristic sequence reflecting the evolution of the motor's operating state. Transforming discrete data points into multiple water flow sequences and multiple motor characteristic sequences with temporal characteristics provides a structured data foundation for subsequent analysis.

[0016] Finally, during the production of all-cotton spunlace nonwoven fabric, images are captured of multiple continuous areas after spunlace treatment along the fiber web's forward direction. Specifically, if a spunlace component malfunctions, insufficient fiber web entanglement at the corresponding location will result in a potential defect area. When this area is conveyed to a subsequent normal spunlace component along with the fiber web, the high-pressure water jets will preferentially penetrate the looser parts of the structure, causing some fibers to be dispersed and carried away by the water flow, while other fibers are squeezed to both sides of the defect area, thus forming gradually aggravated streak defects.

[0017] Image acquisition devices deployed on the production line, such as industrial line scan cameras or high-resolution area scan cameras, continuously capture images of the fiber web surface covering different production moments and different areas of the web. Arranging and combining the acquired images according to time sequence and spatial location creates a fiber web image sequence that comprehensively reflects the continuous changes in the surface morphology and structural characteristics of the fiber web throughout the entire production process. This fiber web image sequence clearly records the evolution of the fiber web's apparent state after hydroentangling, providing an intuitive and reliable image evidence basis for subsequent automatic analysis of streak defects and intelligent identification of equipment faults based on computer vision.

[0018] S20: Input each fiber web image in the fiber web image sequence into the stre recognition network group, output the stre defect rate sequence, and process it to obtain the stre hidden failure rate; Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and a streak defect rate sequence is output. Processing is then performed to obtain the streak latent failure rate, including: Invoke the pre-configured streak recognition network group; Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and multiple defect recognition result sets are output, wherein each defect recognition result includes whether a defect exists or not. The proportion of defects in multiple defect identification result sets is calculated separately and used as the streak defect rate to obtain the streak defect rate sequence; Based on the streak defect rate sequence, the streak hidden failure rate is obtained.

[0019] First, a pre-configured streak recognition network is invoked. This network consists of multiple deep learning models trained on a large number of historical fiber mesh image samples. It has a high-precision ability to identify streak defects in fiber mesh images and can perform defect existence judgments in parallel for each input fiber mesh image, and output the recognition conclusion.

[0020] Specifically, the pre-configuration steps of the streak recognition network group include: Based on historical production data of all-cotton spunlace production, a set of sample fiber web images was collected, and the presence of streak defects in each sample fiber web image was marked to obtain a set of sample defect identification results. Based on deep learning, multiple streak recognition networks with the same architecture were constructed. Training and testing data are randomly extracted multiple times from the sample fiber web image set and sample defect identification result set. Each streak recognition network is trained and tested. After all tests converge, a streak recognition network group is obtained and configured in the control console of the cotton spunlace production workshop.

[0021] First, training samples are prepared and labeled. Specifically, based on historical production data accumulated during the cotton spunlace production process, fiber web images representing different production states and defect types are collected to form an initial sample fiber web image set. Historical production data refers to equipment operating parameters, process settings, and corresponding product quality reports recorded in past production cycles, obtained through the production line database system and image archiving system. Then, professional technicians, according to unified defect judgment standards, conduct precise manual review and labeling of each sample fiber web image in the sample fiber web image set, clearly identifying whether streak defects exist, generating a sample defect identification result set corresponding one-to-one with the sample fiber web image set. The final sample fiber web image set comprehensively covers the fiber web appearance characteristics under normal production conditions and various abnormal operating conditions, while the sample defect identification result set provides authoritative defect annotations for each image. Both together provide an accurate data foundation for the subsequent training of the deep learning model.

[0022] Secondly, based on a deep learning framework, multiple stripe recognition networks with the same network structure, initialization parameters, and connection methods were designed and constructed. Deep learning is a machine learning method that automatically learns feature representations from data through multi-layer nonlinear transformations, including various typical architectures such as convolutional neural networks and recurrent neural networks. The constructed stripe recognition networks adopt a unified architecture, ensuring benchmark consistency when different stripe recognition networks process the same image recognition task.

[0023] Finally, the model is trained and deployed. Specifically, a random sampling method is used to independently divide the prepared sample fiber web image set and sample defect recognition result set into training and test subsets multiple times. Each constructed streak recognition network is trained and tested independently using different data partitions. Model parameters are optimized through forward propagation, loss calculation, and backpropagation until all streak recognition networks exhibit stable performance on their respective test sets and reach convergence. The convergence condition is set according to a preset training termination criterion, such as the validation set loss function value no longer significantly decreasing after 20 consecutive training cycles, or the classification accuracy on the validation set reaching a predetermined threshold of 95%. Finally, the fully trained and qualified streak recognition network group is integrated into the control computing unit of the cotton spunlace production workshop to complete the deployment of online recognition capabilities.

[0024] For example, since there is a complex nonlinear mapping relationship between the streak defect features in the fiber mesh image and the equipment failure state, and convolutional neural networks have outstanding advantages in image feature extraction and pattern recognition, convolutional neural networks can be selected as the basic architecture of the streak recognition network.

[0025] Specifically, this streak recognition network mainly consists of a feature extraction module, a spatial attention module, and a classification output module. The feature extraction module employs a deep convolutional neural network structure, containing five convolutional layers and three max-pooling layers. The kernel size varies from 3×3 to 7×7 depending on the image resolution, and a ReLU activation function is used after each convolutional layer to enhance non-linear expressiveness. The spatial attention module adaptively adjusts the weights of each channel of the feature map through a feature recalibration mechanism, increasing attention to the streak region. The classification output module uses a global average pooling layer and two fully connected layers, ultimately outputting the probability of defect presence through a sigmoid activation function.

[0026] During network training, key hyperparameters included an initial learning rate of 0.001, 200 training epochs, and a batch size of 32. The learning rate was set dynamically using a cosine annealing strategy based on the characteristics of the image classification task. The number of training epochs ensured the model fully learned the features of various scratch defects, and the batch size balanced memory utilization and training stability. Specifically, supervised learning was used, with a set of sample fiber mesh images collected from historical production data as the input sample set and the corresponding set of sample defect identification results as the label set. The sample dataset was randomly divided into training, validation, and test sets in a 6:2:2 ratio to ensure a balanced distribution of normal and defective samples in each set.

[0027] Furthermore, fiber mesh image samples from the training set are used as input, with corresponding defect labels as supervision signals. The network parameters are updated using a backpropagation algorithm combined with an Adam optimizer. A binary cross-entropy loss function is employed to measure the difference between the predicted results and the true labels, and the training process is monitored using validation set accuracy. Training is terminated when the validation set accuracy no longer improves and reaches above 95% for 20 consecutive training epochs, resulting in a stable streak recognition network. This training process is repeated to construct multiple streak recognition networks with identical structures but different initialization parameters, ultimately forming a streak recognition network ensemble. This streak recognition network ensemble can effectively identify various streak defects in fiber mesh images, providing a reliable visual analysis foundation for equipment fault diagnosis.

[0028] Furthermore, each fiber web image in the fiber web image sequence is sequentially input into the streak recognition network for inference and analysis. The streak recognition network independently judges each input fiber web image and outputs the corresponding defect recognition result. This defect recognition result is a binary judgment, explicitly indicating whether a defect exists or not in the fiber web image. When the entire fiber web image sequence has been processed, multiple defect recognition results corresponding to the number of images in the fiber web image sequence will be obtained, forming multiple defect recognition result sets.

[0029] Furthermore, for each preset analysis time window or image acquisition batch, the number of fiber web images identified as having streak defects by the streak recognition network within that time window or batch is counted, and the percentage of such images in the total number of fiber web images in that time window or batch is calculated. This percentage is the streak defect rate for the corresponding analysis time window or image batch. The preset analysis time window is set according to the production cycle and the timeliness requirements of fault diagnosis, for example, a statistical cycle of 5 minutes. The preset image acquisition batch refers to a group of images divided according to a fixed number of acquisitions, determined based on the camera acquisition frequency and statistical significance requirements; for example, every 100 consecutively acquired fiber web images are divided into an independent batch for statistical analysis. The final streak defect rate is a quantitative indicator characterizing the degree of quality deterioration of fiber web products within a specific time period or production unit. Its value directly reflects the production stability and equipment operating status during the corresponding period.

[0030] As the production process of all-cotton spunlace nonwoven fabric continues, the streak defect rate is continuously calculated and recorded in chronological order for each analysis time window. The sequentially arranged streak defect rate values ​​together constitute a complete streak defect rate sequence. This streak defect rate sequence dynamically reveals, in a quantitative form, the probability of streak defects occurring during the production process and their fluctuation patterns over time, providing a time-series data foundation for subsequent fault feature extraction.

[0031] Finally, the streak defect rate sequence is subjected to in-depth processing to extract a more representative streak hidden failure rate. This streak hidden failure rate is a composite index obtained by analyzing the time-series variation characteristics of the streak defect rate. It characterizes the severity and development trend of potential failure states of production equipment, and can effectively distinguish between random quality fluctuations and systemic equipment degradation, providing a key decision-making basis for predictive maintenance.

[0032] Specifically, based on the streak defect rate sequence, the streak implicit failure rate is obtained through processing, including: Calculate the difference between the first and second streak defect rates within the streak defect rate sequence to obtain the first streak anomaly gradient, and then calculate multiple streak anomaly gradients. The mean of the multiple streak anomaly gradients is calculated to obtain the streak hidden failure rate.

[0033] First, in the streak defect rate sequence, two streak defect rate values ​​corresponding to adjacent time points are selected, namely the first streak defect rate and the immediately following second streak defect rate in the sequence, and the numerical difference between them is calculated. This difference is the first streak anomaly gradient, and the magnitude and positive / negative characteristics of the first streak anomaly gradient reflect the variation range and evolution direction of the streak defect rate during continuous production. Following this calculation method, each pair of adjacent streak defect rate values ​​in the streak defect rate sequence is processed sequentially along the time axis, and their corresponding differences are calculated respectively. Finally, a set of multiple streak anomaly gradients arranged in chronological order can be obtained.

[0034] Secondly, a central tendency quantitative analysis is performed on the calculated streak anomaly gradients. Specifically, by calculating the arithmetic mean of all streak anomaly gradients, the discrete gradient observations are integrated into a representative comprehensive index, which is defined as the streak implied failure rate. This data processing method based on the mean algorithm can effectively filter out short-term fluctuations caused by accidental factors in the production process, thereby ensuring that the streak implied failure rate more reliably represents the overall evolution of the streak defect rate within the observation period, and more accurately reveals the continuous development characteristics of potential fault states in cotton spunlace machines.

[0035] The final obtained streak hidden failure rate is a feature quantity processed by gradient analysis and mean fusion, which can effectively quantify the correlation between the evolution trend of streak defects and equipment status, thus providing a reliable quantitative basis for early fault warning and predictive maintenance of cotton hydroentanglement machines.

[0036] S30: Calculate the water intrusion hidden anomaly rate based on the multiple water flow sequences and multiple motor characteristic sequences; Specifically, based on the multiple water flow sequences and multiple motor characteristic sequences, the water intrusion hidden anomaly rate is calculated, including: Within each water flow sequence, the minimum water flow value is selected, and the deviation from the average water flow value is calculated to obtain multiple abnormal water flow values. Calculate the average of the multiple abnormal water flow amplitudes to obtain the average abnormal water flow amplitude; The average amplitude of motor characteristic anomalies is calculated based on multiple motor characteristic sequences. The implied anomaly rate of water jet is calculated based on the abnormal amplitude of the average water flow rate and the abnormal amplitude of the average motor characteristics.

[0037] First, anomaly analysis of water flow is performed. For each hydroentanglement component, the minimum water flow value within the observation period is identified, and the absolute deviation between this minimum value and the arithmetic mean of the water flow sequence is calculated. This deviation value represents the magnitude of the water flow anomaly that characterizes the stability of the water supply to the hydroentanglement component.

[0038] For example, suppose the average water flow rate of a certain hydroentanglement component is 150 liters per minute, and the minimum monitored value is 135 liters per minute. Then, the calculation process for its abnormal water flow rate is 150-135=15. The larger the abnormal water flow rate, the more significant the water flow fluctuation of the hydroentanglement component, and the worse its operational stability.

[0039] Then, the arithmetic mean of the abnormal water flow rate of all hydroentanglement components is calculated to obtain the average abnormal water flow rate, which characterizes the stability of the water supply of the entire hydroentanglement system.

[0040] Simultaneously, motor characteristic anomaly analysis is performed. Based on multiple motor characteristic sequences, the absolute deviation between the minimum value of each motor characteristic sequence and the arithmetic mean of the sequences is calculated as the motor characteristic anomaly amplitude. To comprehensively evaluate the overall state of the drive system, the anomaly amplitudes of different dimensions need to be dimensionless. The arithmetic mean of the dimensionless anomaly amplitudes of all motor characteristics is then calculated to obtain the average motor characteristic anomaly amplitude reflecting the overall operating state of the drive system.

[0041] For example, assuming the mean current characteristic sequence of the drive motor is 28.5 amps and the minimum monitored value is 26.0 amps, then its current characteristic anomaly amplitude is 28.5 - 26.0 = 2.5 amps. Assuming the mean vibration velocity characteristic sequence of the same drive motor is 4.2 mm / s and the minimum monitored value is 3.8 mm / s, then its vibration characteristic anomaly amplitude is 4.2 - 3.8 = 0.4 mm / s. The dimensionless value of the current anomaly amplitude = current characteristic anomaly amplitude / mean current characteristic sequence = 2.5 / 28.5 ≈ 0.088; the dimensionless value of the vibration anomaly amplitude = vibration characteristic anomaly amplitude / mean vibration characteristic sequence = 0.4 / 4.2 ≈ 0.095. The average motor characteristic anomaly amplitude is (0.088 + 0.095) / 2 ≈ 0.092. The larger this average motor characteristic anomaly amplitude value, the more significant the fluctuation in the overall operating state of the drive system relative to its normal baseline level, and the worse its stability.

[0042] Finally, multi-source information fusion is performed. The average water flow anomaly amplitude and the average motor characteristic anomaly amplitude are combined using a weighted summation algorithm to generate a unified spunlace implied anomaly rate. To eliminate the influence of differences in dimensions and numerical magnitudes on the fusion results, the two average anomaly amplitudes are normalized before calculation, converting them into ratios relative to their respective preset benchmark values. This spunlace implied anomaly rate comprehensively reflects the degree of operational anomalies of the spunlace unit in two key dimensions: the hydraulic system and the drive system, providing a quantitative basis for equipment condition assessment.

[0043] Specifically, the hidden anomaly rate of spunlace press is calculated as follows: α × (Average water flow anomaly amplitude / Water flow anomaly baseline) + β × (Average motor characteristic anomaly amplitude / Motor characteristic anomaly baseline). The water flow anomaly baseline is a reference value determined based on historical normal operation data. It is typically the high end of the water flow anomaly amplitude during long-term operation of the spunlace press under fault-free conditions, such as the 95th percentile or the maximum value. Exceeding this baseline indicates a significant deviation from the normal fluctuation range. The motor characteristic anomaly baseline is also determined based on historical normal operation data. Since the average motor characteristic anomaly amplitude is a dimensionless value after dimension normalization, its baseline is set to 1. The weighting coefficients α and β are set according to the contribution of the hydraulic system and drive system to the fault impact in the spunlace press. For example, through regression analysis of historical fault data, the weighting coefficient α for the hydraulic system is determined to be 0.6, and the weighting coefficient β for the drive system is 0.4. The weighting coefficients must satisfy the condition α + β = 1 to ensure that the hidden anomaly rate of spunlace press has consistent dimensions and comparability.

[0044] For example, suppose the average water flow anomaly is 15 liters per minute, and the baseline value for water flow anomaly is 25 liters per minute; the average motor characteristic anomaly is 0.092, which is a dimensionless result calculated based on the relative deviation of current and vibration characteristics, so the baseline value for motor characteristic anomaly is taken as 1. The normalized calculation process is as follows: water flow anomaly ratio is 15 / 25 = 0.6; motor characteristic anomaly ratio is 0.092 / 1 = 0.092. The implied anomaly rate of the spunlace turbine is 0.6 × 0.6 + 0.4 × 0.092 = 0.3968. This implied anomaly rate of the spunlace turbine is a dimensionless value, and its magnitude directly characterizes the degree of anomaly in the overall operating status of the spunlace turbine unit.

[0045] S40: Verify the equipment failure rate based on the implied anomaly rate and the implied failure rate of the spunlace and streak, and use it as the identification result.

[0046] Specifically, the equipment failure rate is obtained by verifying the implied anomaly rate and the implied failure rate of the spunlace layer and streak, and is used as the identification result, including: Based on the implied anomaly rate of the spunlace, the first equipment failure rate is obtained through verification processing. Based on the implied failure rate of the streak, the verification process is used to obtain the second equipment failure rate; The equipment failure rate is calculated based on the first equipment failure rate and the second equipment failure rate, and is used as the identification result.

[0047] First, based on the implied anomaly rate of the spunlace material, the verification process obtains the first equipment failure rate, including: The average hidden anomaly rate of spunlace when a cotton spunlace machine experiences spunlace failure is obtained and used as the baseline hidden anomaly rate of spunlace. The similarity between the implied anomaly rate of the spunlace and the benchmark implied anomaly rate is calculated and used as the first equipment failure rate.

[0048] Specifically, the implicit anomaly rate of spunlace is verified and transformed, converting the implicit anomaly rate of spunlace, which characterizes the degree of abnormality of equipment operating parameters, into an indicator characterizing the probability of equipment failure, namely the first equipment failure rate.

[0049] First, from the historical fault data of the cotton hydroentanglement machine, multiple sets of hidden anomaly rates recorded when hydroentanglement faults were confirmed in the past were extracted, and the arithmetic mean was calculated. This average hidden anomaly rate was determined as the baseline hidden anomaly rate. This baseline hidden anomaly rate represents the typical level of hidden anomaly rate of hydroentanglement equipment under a clear fault condition.

[0050] Secondly, the implied anomaly rate of the spunlace pipe obtained in the current monitoring period is compared with the baseline implied anomaly rate to quantify the equipment failure risk. Specifically, the relative deviation of the current implied anomaly rate of the spunlace pipe relative to the baseline implied anomaly rate of the spunlace pipe is calculated, and the similarity between the two is defined based on this deviation as the first equipment failure rate. This similarity is calculated using the following formula: First equipment failure rate = max(0, 1 - |(current implied anomaly rate of the spunlace pipe - baseline implied anomaly rate of the spunlace pipe) / baseline implied anomaly rate of the spunlace pipe|).

[0051] The first equipment failure rate is a dimensionless numerical indicator between 0 and 1, and its value directly represents the similarity between the equipment state and historical failure states. When the current implied anomaly rate of the spunlace material equals the baseline implied anomaly rate, the relative deviation is 0, and the first equipment failure rate is 1, indicating that the current state is completely similar to the historical failure state. When the current implied anomaly rate of the spunlace material deviates from the baseline value, the absolute value of the relative deviation increases, and the first equipment failure rate decreases from 1. The smaller the value, the greater the difference between the two states and the lower the similarity. This similarity measurement method based on relative deviation achieves a smooth mapping from the magnitude of physical parameter anomalies to the probability of equipment failure, providing an intuitive basis for maintenance decisions.

[0052] Furthermore, the implicit failure rate of the streak is verified and converted into a second equipment failure rate. This conversion process employs a quantization process consistent with the method used to generate the first equipment failure rate.

[0053] Specifically, the second equipment failure rate is calculated as max(0, 1 - |(current streak implied failure rate - baseline streak implied failure rate) / baseline streak implied failure rate|). The baseline streak implied failure rate is obtained by statistically analyzing the average streak implied failure rate corresponding to confirmed equipment failures in historical data. This second equipment failure rate is also a dimensionless numerical indicator between 0 and 1, reflecting the similarity between the current trend of product quality defects and historical failure characteristics.

[0054] Finally, a fusion decision is made based on the two failure probability indicators mentioned above. The failure rates of the first and second equipment are calculated using a weighted summation algorithm to generate a unified, comprehensive equipment failure rate. Specifically, the equipment failure rate = γ × first equipment failure rate + δ × second equipment failure rate. Here, the weighting coefficients γ and δ are real numbers greater than 0, and their specific values ​​can be set according to the operating characteristics of the spunlace unit or expert experience, and can satisfy the normalization condition γ + δ = 1. For example, when the equipment operating parameters are considered to have higher reliability than image features, γ can be set to 0.6 and δ to 0.4.

[0055] The final equipment failure rate takes into account information from two dimensions: the equipment's own operating status and product quality performance. This overcomes the limitations of a single information source and, as the final output of the fault identification result, provides a more accurate and reliable basis for predictive maintenance of the equipment.

[0056] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first constructs a dual-sensor system—combining direct equipment condition monitoring and indirect product quality characterization—by collecting multimodal operating data and fiber web image sequences from the spunlace machine unit, providing a more comprehensive data foundation for fault diagnosis. Secondly, it utilizes a streak recognition network to intelligently analyze the fiber web images, accurately extracting the streak-related hidden fault rate from product surface defects, reflecting the equipment's health status, thus achieving a reverse mapping from quality performance to equipment status. Thirdly, it calculates the spunlace-related hidden anomaly rate based on the water flow rate and motor characteristic sequences of the multi-spunlace components, enabling direct capture of abnormal signs at the equipment operation level. Finally, by converting the spunlace-related hidden anomaly rate and the streak-related hidden fault rate into comparable fault probabilities and performing weighted fusion, it achieves cross-validation and comprehensive decision-making between equipment operating parameters and product quality characteristics. This effectively overcomes the limitations and lag of single-signal-source analysis, improving the accuracy, timeliness, and reliability of fault identification, and providing unified technical support for predictive maintenance and product quality control of all-cotton spunlace machines.

[0057] Example 2, as Figure 2As shown, based on the same inventive concept as the multimodal data fusion-based fault identification method for cotton spunlace machines provided in Embodiment 1, this embodiment of the invention also provides a multimodal data fusion-based fault identification system for cotton spunlace machines, including: The data acquisition module 11 is used to acquire multimodal data of multiple hydroentanglement components during the operation of the all-cotton hydroentanglement machine, obtain multiple water flow sequences and multiple motor characteristic sequences, and acquire fiber web image sequences of all-cotton hydroentangled nonwoven fabric. The streak analysis module 12 is used to input each fiber web image in the fiber web image sequence into the streak recognition network group, output the streak defect rate sequence, and process it to obtain the streak hidden failure rate. The water jet state analysis module 13 is used to calculate the water jet hidden anomaly rate based on the multiple water flow sequences and multiple motor characteristic sequences; The fault identification module 14 is used to verify the equipment failure rate based on the implied anomaly rate of the spunlace and the implied failure rate of the streak, and to obtain the equipment failure rate as the identification result.

[0058] Specifically, the data acquisition module 11 is used for: Multimodal data of multiple hydrospunlace components during the operation of the all-cotton hydrospunlace machine were collected to obtain multiple water flow sequences and multiple motor characteristic sequences. Additionally, a sequence of fiber web images of the all-cotton hydrospunlace nonwoven fabric was acquired, including: Multimodal data of multiple hydrospunlace components were collected during the operation of the all-cotton hydrospunlace machine. The multimodal data included water flow rate and motor characteristics. By integrating multimodal data from multiple hydroentanglement components according to time sequence, multiple water flow sequences and multiple motor characteristic sequences are obtained; Images of multiple web regions along the forward direction during the production of all-cotton spunlace nonwoven fabric were collected to obtain a web image sequence.

[0059] Specifically, the streak analysis module 12 is used for: Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and a streak defect rate sequence is output. Processing is then performed to obtain the streak latent failure rate, including: Invoke the pre-configured streak recognition network group; Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and multiple defect recognition result sets are output, wherein each defect recognition result includes whether a defect exists or not. The proportion of defects in multiple defect identification result sets is calculated separately and used as the streak defect rate to obtain the streak defect rate sequence; Based on the streak defect rate sequence, the streak hidden failure rate is obtained.

[0060] Specifically, the pre-configuration steps of the streak recognition network group include: Based on historical production data of all-cotton spunlace production, a set of sample fiber web images was collected, and the presence of streak defects in each sample fiber web image was marked to obtain a set of sample defect identification results. Based on deep learning, multiple streak recognition networks with the same architecture were constructed. Training and testing data are randomly extracted multiple times from the sample fiber web image set and sample defect identification result set. Each streak recognition network is trained and tested. After all tests converge, a streak recognition network group is obtained and configured in the control console of the cotton spunlace production workshop.

[0061] Specifically, based on the streak defect rate sequence, the streak implicit failure rate is obtained through processing, including: Calculate the difference between the first and second streak defect rates within the streak defect rate sequence to obtain the first streak anomaly gradient, and then calculate multiple streak anomaly gradients. The mean of the multiple streak anomaly gradients is calculated to obtain the streak hidden failure rate.

[0062] The hydroentanglement state analysis module 13 is specifically used for: Based on the multiple water flow sequences and multiple motor characteristic sequences, the hidden anomaly rate of water jetting is calculated, including: Within each water flow sequence, the minimum water flow value is selected, and the deviation from the average water flow value is calculated to obtain multiple abnormal water flow values. Calculate the average of the multiple abnormal water flow amplitudes to obtain the average abnormal water flow amplitude; The average amplitude of motor characteristic anomalies is calculated based on multiple motor characteristic sequences. The implied anomaly rate of water jet is calculated based on the abnormal amplitude of the average water flow rate and the abnormal amplitude of the average motor characteristics.

[0063] The fault identification module 14 is specifically used for: The equipment failure rate is obtained by verifying the implied anomaly rate and streak implied failure rate of the spunlace material, and is used as the identification result, including: Based on the implied anomaly rate of the spunlace, the first equipment failure rate is obtained through verification processing. Based on the implied failure rate of the streak, the verification process is used to obtain the second equipment failure rate; The equipment failure rate is calculated based on the first equipment failure rate and the second equipment failure rate, and is used as the identification result.

[0064] Specifically, based on the implied anomaly rate of the spunlace material, the verification process obtains the first equipment failure rate, including: The average hidden anomaly rate of spunlace when a cotton spunlace machine experiences spunlace failure is obtained and used as the baseline hidden anomaly rate of spunlace. The similarity between the implied anomaly rate of the spunlace and the benchmark implied anomaly rate is calculated and used as the first equipment failure rate.

[0065] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0066] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0067] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for fault identification of a cotton hydroentanglement machine based on multimodal data fusion, characterized in that, The method includes: Multimodal data of multiple hydrospun parts during the operation of the all-cotton hydrospun machine were collected to obtain multiple water flow sequences and multiple motor characteristic sequences, and the fiber web image sequence of the all-cotton hydrospun nonwoven fabric was also collected. Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and a streak defect rate sequence is output. The streak hidden failure rate is then obtained through processing. Based on the multiple water flow sequences and multiple motor characteristic sequences, the water jet hidden anomaly rate is calculated; The equipment failure rate is obtained by verifying the implied anomaly rate and the implied failure rate of the spunlace and streak, and is used as the identification result.

2. The multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines according to claim 1, characterized in that, Multimodal data of multiple hydrospunlace components during the operation of the all-cotton hydrospunlace machine were collected to obtain multiple water flow sequences and multiple motor characteristic sequences. Additionally, a sequence of fiber web images of the all-cotton hydrospunlace nonwoven fabric was acquired, including: Multimodal data of multiple hydrospunlace components were collected during the operation of the all-cotton hydrospunlace machine. The multimodal data included water flow rate and motor characteristics. By integrating multimodal data from multiple hydroentanglement components according to time sequence, multiple water flow sequences and multiple motor characteristic sequences are obtained; Images of multiple web regions along the forward direction during the production of all-cotton spunlace nonwoven fabric were collected to obtain a web image sequence.

3. The method for fault identification of a cotton hydroentanglement machine based on multimodal data fusion according to claim 1, characterized in that, Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and a streak defect rate sequence is output. Processing is then performed to obtain the streak latent failure rate, including: Invoke the pre-configured streak recognition network group; Each fiber web image in the fiber web image sequence is input into a streak recognition network group, and multiple defect recognition result sets are output, wherein each defect recognition result includes whether a defect exists or not. The proportion of defects in multiple defect identification result sets is calculated separately and used as the streak defect rate to obtain the streak defect rate sequence; Based on the streak defect rate sequence, the streak hidden failure rate is obtained.

4. The multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines according to claim 3, characterized in that, The pre-configuration steps for the streak recognition network group include: Based on historical production data of all-cotton spunlace production, a set of sample fiber web images was collected, and the presence of streak defects in each sample fiber web image was marked to obtain a set of sample defect identification results. Based on deep learning, multiple streak recognition networks with the same architecture were constructed. Training and testing data are randomly extracted multiple times from the sample fiber web image set and sample defect identification result set. Each streak recognition network is trained and tested. After all tests converge, a streak recognition network group is obtained and configured in the control console of the cotton spunlace production workshop.

5. The multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines according to claim 3, characterized in that, Based on the streak defect rate sequence, the streak implicit failure rate is obtained through processing, including: Calculate the difference between the first and second streak defect rates within the streak defect rate sequence to obtain the first streak anomaly gradient, and then calculate multiple streak anomaly gradients. The mean of the multiple streak anomaly gradients is calculated to obtain the streak hidden failure rate.

6. The method for fault identification of a cotton hydroentanglement machine based on multimodal data fusion according to claim 1, characterized in that, Based on the multiple water flow sequences and multiple motor characteristic sequences, the hidden anomaly rate of water jetting is calculated, including: Within each water flow sequence, the minimum water flow value is selected, and the deviation from the average water flow value is calculated to obtain multiple abnormal water flow values. Calculate the average of the multiple abnormal water flow rates to obtain the average abnormal water flow rate. The average amplitude of motor characteristic anomalies is calculated based on multiple motor characteristic sequences. The implied anomaly rate of water jet is calculated based on the abnormal amplitude of the average water flow rate and the abnormal amplitude of the average motor characteristics.

7. The multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines according to claim 1, characterized in that, The equipment failure rate is obtained by verifying the implied anomaly rate and streak implied failure rate of the spunlace material, and is used as the identification result, including: Based on the implied anomaly rate of the spunlace, the first equipment failure rate is obtained through verification processing. Based on the implied failure rate of the streak, the verification process is used to obtain the second equipment failure rate; The equipment failure rate is calculated based on the first equipment failure rate and the second equipment failure rate, and is used as the identification result.

8. The multimodal data fusion-based fault identification method for all-cotton hydroentanglement machines according to claim 7, characterized in that, Based on the implied anomaly rate of the spunlace material, the verification process obtains the first equipment failure rate, including: The average hidden anomaly rate of spunlace when a cotton spunlace machine experiences spunlace failure is obtained and used as the baseline hidden anomaly rate of spunlace. The similarity between the implied anomaly rate of the spunlace and the benchmark implied anomaly rate is calculated and used as the first equipment failure rate.

9. A multimodal data fusion-based fault identification system for all-cotton hydrospunlace machines, characterized in that, The method for fault identification of a cotton spunlace machine for performing multimodal data fusion as described in any one of claims 1-8 includes: The data acquisition module is used to collect multimodal data of multiple hydroentanglement components during the operation of the all-cotton hydroentanglement machine, obtain multiple water flow sequences and multiple motor characteristic sequences, and acquire fiber web image sequences of all-cotton hydroentangled nonwoven fabric. The streak analysis module is used to input each fiber web image in the fiber web image sequence into the streak recognition network group, output the streak defect rate sequence, and process it to obtain the streak hidden failure rate. The hydroentanglement status analysis module is used to calculate the hydroentanglement hidden anomaly rate based on the multiple water flow sequences and multiple motor characteristic sequences; The fault identification module is used to verify the equipment failure rate based on the implied anomaly rate of the spunlace and the implied failure rate of the streak, and to obtain the equipment failure rate as the identification result.

Images