Edge computing system and method for clothing flexible production line intelligent agent equipment data acquisition

Abstract

The invention discloses an edge computing system and method for clothing flexible production line intelligent agent equipment data acquisition, and belongs to the technical field of clothing production automation and intelligence. The system adopts a three-layer architecture of an intelligent agent equipment layer, an edge computing layer and a cloud management layer, equipment operation data is acquired through a sensor, preprocessing such as data cleaning, noise reduction and normalization and feature extraction are performed on an edge side, and local decision is realized by combining rule decision and a lightweight machine learning model. And deep analysis and feedback are carried out by utilizing a deep learning model through cooperation of incremental transmission and timed transmission strategies and a cloud end. According to the scheme, the real-time performance and efficiency of data processing are improved, the network delay and the operation cost are reduced, the system reliability is enhanced, and the method is suitable for intelligent upgrading of the clothing flexible production line.

Description

Technical Field

[0001] This invention relates to the field of automation and intelligent technology in garment production, specifically to an edge computing system and method for data acquisition of intelligent equipment in flexible garment production lines. Through edge computing and cloud collaboration mechanisms, it achieves efficient processing of production data and intelligent decision-making, and is applicable to the intelligent upgrade scenarios of flexible garment production lines. Background Technology

[0002] Traditional flexible garment production lines suffer from significant drawbacks in data acquisition and processing. In a centralized architecture, all equipment data must be transmitted to a central server for processing, resulting in massive data transfer volumes and network congestion. Simultaneous data transmission from sewing machines and cutting tables can lead to network latency of hundreds of milliseconds, failing to meet real-time control requirements. Furthermore, the data processing chain is long, requiring multiple transmission stages from acquisition to feedback. For example, abnormal ironing temperatures can cause fabric damage due to central server processing delays. The system's reliability is overly reliant on the central server; a server failure can paralyze the entire production line's data acquisition and processing system, with an average recovery time exceeding two hours. Current edge computing technology in the garment industry also has significant limitations. Its multi-source data processing capabilities are insufficient, failing to effectively integrate heterogeneous data such as sewing machine speed, cutting table pressure, and ironing humidity, resulting in feature extraction accuracy below 70%. Moreover, the edge-cloud collaboration mechanism is inefficient, with traditional timed transmission strategies leading to over 60% redundant data transmission and cloud-based deep analysis results experiencing delays exceeding 10 minutes. Summary of the Invention

[0003] The purpose of this invention is to provide an edge computing system and method for data acquisition of intelligent body equipment in flexible garment production lines. By performing data preprocessing, feature extraction, and local decision-making on edge devices, the system reduces data transmission volume, lowers network latency, and improves data processing efficiency. At the same time, through an edge-cloud collaborative mechanism, it enables in-depth data analysis and global decision-making, supporting the intelligent operation of flexible garment production lines.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an edge computing system for data acquisition of intelligent body equipment in a flexible garment production line, comprising:

[0005] The intelligent device layer includes various devices and supporting sensors on the flexible garment production line, used to collect equipment operation data;

[0006] The edge computing layer, deployed in the area surrounding the intelligent agent device layer, consists of edge computing nodes used to receive and process data and make local decisions.

[0007] The cloud management layer is used to perform in-depth analysis, storage, and global decision-making on the data uploaded from the edge computing layer, and then feed it back to the edge computing layer.

[0008] After the sensors in the intelligent agent device layer collect device operation data according to the set frequency and format, they transmit the data in real time to the edge computing nodes in the edge computing layer through the industrial communication protocol. After receiving the data and completing preprocessing, feature extraction and local decision-making, the edge computing nodes generate device control commands and transmit them back to the corresponding devices in the intelligent agent device layer through the original communication protocol to realize real-time control of the devices.

[0009] As a further improvement to the technical solution of the present invention, the sensors of the intelligent device layer include: a speed sensor, a thread tension sensor, and a stitch distance sensor installed on a sewing machine; a position sensor and a pressure sensor installed on a cutting bed; and a temperature sensor and a humidity sensor installed on an ironing device.

[0010] As a further improvement to the technical solution of the present invention, the edge computing node uses an NVIDIA Jetson series development board as the hardware carrier.

[0011] As a further improvement to the technical solution of this invention, an edge computing method for data acquisition of intelligent body equipment in a flexible garment production line includes the following steps:

[0012] Step S1, Data Acquisition and Preprocessing: Sensors in the intelligent agent device layer acquire device operation data according to a set frequency and format, and transmit it to the edge computing node in the edge computing layer through an industrial communication protocol; The edge computing node cleans the received raw data (sets a threshold to remove outliers), reduces noise (using Kalman filtering algorithm), and normalizes it (maps it to the [-1,1] interval) to generate preprocessed data;

[0013] Step S2, Feature Extraction and Local Decision-Making: Edge computing nodes extract features from the preprocessed data to obtain time-domain features (mean, variance, peak value), frequency-domain features (dominant frequency, spectral energy), and time-frequency features (wavelet transform features); based on the extracted features, local decisions are made through a rule engine or a lightweight machine learning model (decision tree, support vector machine) to generate device control commands or data to be transmitted.

[0014] Step S3, Edge-Cloud Collaboration: Edge computing nodes transmit data requiring in-depth analysis in local decision-making to the cloud management layer using an incremental transmission strategy (transmitting only changed or newly added data) and a timed transmission strategy (transmitting in batches at preset time intervals). The cloud management layer performs in-depth analysis on the received data (using convolutional neural networks or recurrent neural networks) and generates global decision results, which are then fed back to the edge computing nodes via the network. After receiving the feedback results, the edge computing nodes combine them with local decision-making logic to perform subsequent operations (such as optimizing device control parameters and updating decision rules), forming a closed-loop data flow.

[0015] As a further improvement to the technical solution of the present invention, in the data acquisition and preprocessing, outliers are removed by setting a threshold, and the Kalman filter algorithm is used to reduce noise in the data, so that data of different dimensions are normalized to the [-1,1] interval.

[0016] As a further improvement to the technical solution of the present invention, the feature extraction includes:

[0017] Temporal feature extraction involves performing sliding window calculations on preprocessed continuous time-series data (such as the thread tension sampling value of a sewing machine within 30 seconds and the real-time pressure change sequence of a cutting bed) to obtain the mean, variance, and peak value of the data, reflecting the overall trend and fluctuation characteristics of the data in the time dimension.

[0018] Frequency domain feature extraction transforms the same set of data (such as the time series data of cutting bed vibration signal) after time domain analysis into the frequency domain through Fourier transform to obtain the main frequency and spectral energy, which are used to identify the periodic features hidden in the data (such as the specific fault frequency corresponding to tool wear).

[0019] Time-frequency feature extraction is used for complex data (such as sewing machine current signals) that contain both time-domain trends and frequency-domain characteristics. Based on the analysis of time-domain and frequency-domain features, wavelet transform is used to perform joint time-frequency domain analysis. At the same time, the variation pattern of the data on the time axis and the distribution characteristics of different frequency components are preserved, so as to achieve multi-dimensional feature fusion of the preprocessed data.

[0020] As a further improvement to the technical solution of the present invention, the extracted feature data is correlated and judged based on preset rules. For example, the time domain mean (reflecting the overall tension level), variance (reflecting tension stability), and frequency domain dominant frequency (reflecting the tension fluctuation frequency) of the sewing machine thread tension are combined. When the mean exceeds the upper limit and the variance is large, it is determined that the thread tension is abnormal and an alarm is triggered.

[0021] Alternatively, lightweight machine learning models such as decision trees and support vector machines can be deployed, using time-domain features (such as mean and peak values) as basic input features, and frequency-domain features (such as spectral energy) and time-frequency features (such as wavelet transform coefficients) as supplementary features. By synergistically inputting multi-dimensional features, the model's decision-making accuracy can be improved, enabling comprehensive evaluation and real-time response to the equipment's operating status.

[0022] As a further improvement to the technical solution of the present invention, the incremental transmission strategy is as follows: when the device data does not change within a set time, it is not transmitted repeatedly, but only the changed or newly added data is transmitted.

[0023] As a further improvement to the technical solution of the present invention, the timed transmission strategy is as follows: the data accumulated by the edge device is packaged and transmitted to the cloud in batches at preset time intervals.

[0024] As a further improvement to the technical solution of the present invention, the cloud-based deep data analysis employs deep learning algorithms such as convolutional neural networks or recurrent neural networks to model and predict the data, and feeds the results back to the edge computing layer.

[0025] In summary, this invention collects device operation data through sensors, performs preprocessing such as data cleaning, noise reduction, and normalization, and extracts features at the edge. It then combines rule-based decision-making and a lightweight machine learning model to achieve local decision-making. Furthermore, it collaborates with the cloud through incremental and timed transmission strategies, utilizing deep learning models for in-depth analysis and feedback. This invention has the following beneficial effects:

[0026] Improved data processing efficiency: Edge devices process data in real time, reducing the amount of data transmitted to the cloud, lowering network latency, and improving the speed and real-time performance of data processing.

[0027] Enhanced system reliability: The edge computing layer offloads some of the computing tasks from the cloud, reducing reliance on the central server and improving system reliability and stability.

[0028] Supports intelligent decision-making: Local decision-making algorithms can respond promptly to anomalies in the production process, while cloud-based in-depth analysis provides a basis for optimizing the production process, supporting the intelligent operation of flexible garment production lines. Attached Figure Description

[0029] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0030] Figure 1 is a system framework diagram of an edge computing system for data acquisition of intelligent body equipment in a flexible garment production line according to an embodiment of the present invention.

[0031] Figure 2 is a data processing flowchart of an edge computing method for data acquisition of intelligent body equipment in a flexible garment production line according to an embodiment of the present invention.

[0032] Figure 3 is a timing diagram of the collaborative mechanism in an embodiment of the present invention. Detailed Implementation

[0033] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

[0034] It should be noted that all directional indicators (such as up, down, left, right, front, back, upper end, lower end, top, bottom, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0035] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0036] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination should be considered non-existent and not within the scope of protection claimed by this invention.

[0037] The present invention will be further described in detail below with reference to the accompanying drawings.

[0038] Reference Figures 1 to 3 ,in, Figure 3 The sequence of incremental transmission (change triggering mechanism) and timed transmission (time polling mechanism) is shown, and the cloud analysis feedback link is marked (downlink delay of decision results ≤120ms).

[0039] An edge computing system for data acquisition by intelligent body equipment in a flexible garment production line includes:

[0040] The intelligent device layer includes various devices and supporting sensors on the flexible garment production line, used to collect equipment operation data;

[0041] The edge computing layer, deployed in the area surrounding the intelligent agent device layer, consists of edge computing nodes used to receive and process data and make local decisions.

[0042] The cloud management layer is used to perform in-depth analysis, storage, and global decision-making on the data uploaded from the edge computing layer, and then feed it back to the edge computing layer.

[0043] After the sensors in the intelligent agent device layer collect device operation data according to the set frequency and format, they transmit the data in real time to the edge computing nodes in the edge computing layer through the industrial communication protocol. After receiving the data and completing preprocessing, feature extraction and local decision-making, the edge computing nodes generate device control commands and transmit them back to the corresponding devices in the intelligent agent device layer through the original communication protocol to realize real-time control of the devices.

[0044] Specifically, in this embodiment, the sensors of the intelligent device layer include: a speed sensor, a thread tension sensor, and a stitch distance sensor installed on the sewing machine; a position sensor and a pressure sensor installed on the cutting bed; and a temperature sensor and a humidity sensor installed on the ironing equipment.

[0045] It should be noted that in a flexible garment production line, sewing machines, cutting tables, and ironing equipment work in sequence according to the production process, forming a continuous garment processing chain:

[0046] First, the cutting table, as the starting point of the process, cuts the fabric according to the design pattern. Position sensors locate the fabric coordinates, and pressure sensors monitor pressure changes during the cutting process (e.g., the pressure of the cutter on the fabric must be maintained within a reasonable range to ensure cutting accuracy), ensuring that the cut fabric size and shape meet process standards, providing qualified cut pieces for subsequent sewing stages.

[0047] After the cutting table completes the cutting, the cut pieces are transferred to the sewing machine. After receiving the cut pieces, the sewing machine uses a speed sensor to monitor the sewing speed in real time (which affects sewing efficiency and stitch uniformity), a thread tension sensor to control the thread tension (to prevent thread breakage or loose stitches), and a stitch distance sensor to adjust the stitch density (to adapt to different fabrics and process requirements). The cut pieces are then sewn together according to the designed stitch lines to form a semi-finished garment piece.

[0048] The sewn garment pieces are then placed into an ironing machine for shaping. The ironing machine uses a temperature sensor to precisely control the ironing temperature (set according to the fabric material, such as different suitable temperatures for cotton and synthetic fabrics), and a humidity sensor to regulate the humidity of the ironing environment (to prevent the fabric from being damaged due to excessive dryness or deformed due to excessive moisture). Through the action of heat and pressure, the garment pieces are flattened and shaped, ultimately forming a qualified finished garment.

[0049] The three components form a closely connected production line through the flow of production materials. The processing quality of the preceding equipment directly affects the operational effect of the following stage. The operating data of each piece of equipment (such as the cutting accuracy of the cutting bed, the sewing parameters of the sewing machine, and the temperature and humidity of the ironing equipment) are collected by sensors and together provide a basis for the intelligent control of the production line.

[0050] Specifically, in this embodiment, the edge computing node uses an NVIDIA Jetson series development board as its hardware carrier.

[0051] Reference Figure 2 It details the entire timeline of data acquisition, from acquisition (sampling frequency control module), preprocessing (cleaning-noise reduction-normalization pipeline), feature extraction (three-domain feature calculation unit) to local decision-making (parallel processing of rule engine and model inference); specifically, in this embodiment, an edge computing method for data acquisition of intelligent body equipment in a flexible garment production line includes the following steps:

[0052] Step S1, Data Acquisition and Preprocessing: Sensors in the intelligent agent device layer acquire device operation data according to a set frequency and format, and transmit it to the edge computing node in the edge computing layer through an industrial communication protocol; The edge computing node cleans the received raw data (sets a threshold to remove outliers), reduces noise (using Kalman filtering algorithm), and normalizes it (maps it to the [-1,1] interval) to generate preprocessed data;

[0053] Step S2, Feature Extraction and Local Decision-Making: Edge computing nodes extract features from the preprocessed data to obtain time-domain features (mean, variance, peak value), frequency-domain features (dominant frequency, spectral energy), and time-frequency features (wavelet transform features); based on the extracted features, local decisions are made through a rule engine or a lightweight machine learning model (decision tree, support vector machine) to generate device control commands or data to be transmitted.

[0054] Step S3, Edge-Cloud Collaboration: Edge computing nodes transmit data requiring in-depth analysis in local decision-making to the cloud management layer using an incremental transmission strategy (transmitting only changed or newly added data) and a timed transmission strategy (transmitting in batches at preset time intervals). The cloud management layer performs in-depth analysis on the received data (using convolutional neural networks or recurrent neural networks) and generates global decision results, which are then fed back to the edge computing nodes via the network. After receiving the feedback results, the edge computing nodes combine them with local decision-making logic to perform subsequent operations (such as optimizing device control parameters and updating decision rules), forming a closed-loop data flow.

[0055] Specifically, in this embodiment, during data acquisition and preprocessing, outliers are removed by setting a threshold, and the Kalman filter algorithm is used to reduce noise in the data, normalizing data of different dimensions to the [-1,1] interval.

[0056] Specifically, in this embodiment, the feature extraction is based on the unified data preprocessed in step S1, and the three are progressively related through data flow, specifically including:

[0057] Temporal feature extraction involves performing sliding window calculations on preprocessed continuous time-series data (such as the thread tension sampling value of a sewing machine within 30 seconds and the real-time pressure change sequence of a cutting bed) to obtain the mean, variance, and peak value of the data, reflecting the overall trend and fluctuation characteristics of the data in the time dimension.

[0058] Frequency domain feature extraction transforms the same set of data (such as the time series data of cutting bed vibration signal) after time domain analysis into the frequency domain through Fourier transform to obtain the main frequency and spectral energy, which are used to identify the periodic features hidden in the data (such as the specific fault frequency corresponding to tool wear).

[0059] Time-frequency feature extraction is used for complex data (such as sewing machine current signals) that contain both time-domain trends and frequency-domain characteristics. Based on the analysis of time-domain and frequency-domain features, wavelet transform is used to perform joint time-frequency domain analysis. At the same time, the variation pattern of the data on the time axis and the distribution characteristics of different frequency components are preserved, so as to achieve multi-dimensional feature fusion of the preprocessed data.

[0060] Specifically, in this embodiment, local decision-making is based on the time-domain, frequency-domain, and time-frequency features extracted in step S2. These three elements form a decision-making chain through feature data flow, specifically including:

[0061] Based on preset rules, the extracted feature data is correlated and judged. For example, the time domain mean (reflecting the overall tension level), variance (reflecting tension stability), and frequency domain dominant frequency (reflecting the tension fluctuation frequency) of the sewing machine thread tension are combined. When the mean exceeds the upper limit and the variance is large, it is judged as abnormal thread tension and an alarm is triggered.

[0062] Alternatively, lightweight machine learning models such as decision trees and support vector machines can be deployed, using time-domain features (such as mean and peak values) as basic input features, and frequency-domain features (such as spectral energy) and time-frequency features (such as wavelet transform coefficients) as supplementary features. By synergistically inputting multi-dimensional features, the model's decision-making accuracy can be improved, enabling comprehensive evaluation and real-time response to the equipment's operating status.

[0063] Specifically, in this embodiment, the incremental transmission strategy is as follows: when the device data does not change within a set time, it is not transmitted repeatedly, but only the changed or newly added data is transmitted.

[0064] Specifically, in this embodiment, the timed transmission strategy is to batch package and transmit the data accumulated by the edge device to the cloud at preset time intervals.

[0065] Specifically, in this embodiment, the cloud-based deep data analysis employs deep learning algorithms such as convolutional neural networks or recurrent neural networks to model and predict the data, and then feeds the results back to the edge device.

[0066] It should be noted that this system adopts a three-tier hierarchical architecture:

[0067] Intelligent Agent Device Layer

[0068] Includes equipment for the entire garment production process:

[0069] Cutting unit: integrates a position sensor (positioning accuracy 0.5mm) and a pressure sensor (measuring range 0-500N).

[0070] Sewing unit: Equipped with a speed sensor (sampling frequency 100Hz), a thread tension sensor (accuracy 0.1N), and a stitch distance sensor (resolution 0.01mm).

[0071] Ironing unit: Equipped with a temperature sensor (measurement range 0-300℃) and a humidity sensor (accuracy ±2% RH).

[0072] Edge computing layer

[0073] The data processing engine is deployed using the NVIDIA Jetson AGX Orin development board as the edge node.

[0074] Real-time preprocessing module: integrates Kalman filtering algorithm (noise suppression ratio ≥20dB) and minimum-maximum normalization algorithm.

[0075] Feature extraction engine: Supports time-domain features (mean and variance calculation error ≤ 0.5%), frequency-domain features (Fourier transform frequency resolution 1Hz), and time-frequency features (wavelet transform scale parameters 1-32).

[0076] Local decision unit: built-in rule engine (decision latency ≤10ms) and lightweight SVM model (classification accuracy ≥95%).

[0077] It should be noted that:

[0078] 1. Decision-making steps of the built-in rule engine

[0079] The built-in rule engine analyzes and executes decisions on the feature-extracted data based on preset production process standards and equipment operating thresholds. The specific steps are as follows:

[0080] (1) Rule preset and storage

[0081] Based on garment manufacturing process requirements (such as sewing parameters for different fabrics, cutting precision standards, and ironing temperature and humidity ranges), multi-dimensional decision rules are preset and stored in the rule base of the edge computing nodes. For example:

[0082] Sewing machine rules: Upper limit of average thread tension (e.g., 30N), upper limit of variance (e.g., 5N²), an alarm is triggered when both exceed the limit; Speed ​​fluctuation range (e.g., ±5% of rated speed), parameter adjustment is triggered when the range is exceeded.

[0083] Cutting bed rules: average pressure sensor value (e.g., 200-300N), position deviation threshold (e.g., ±0.5mm). If either exceeds the standard, the cutting is judged as abnormal.

[0084] Ironing equipment rules: temperature threshold (e.g., 150-180℃ for cotton fabrics), humidity threshold (e.g., 40%-60%), and cooling or humidity adjustment commands will be activated when the limits are exceeded.

[0085] (2) Feature data input and parsing

[0086] Receive the time-domain, frequency-domain, and time-frequency feature data extracted in step S2 (such as the mean and variance of the sewing machine's thread tension over 30 seconds, and the frequency-domain main frequency of the cutting bed pressure), parse them according to the equipment type, and match them with the rule dimensions of the corresponding equipment in the rule base.

[0087] (3) Rule matching and condition judgment

[0088] The rule engine matches the input feature data one by one according to preset logic (such as "AND / OR" logic):

[0089] Single-condition matching: If the time-domain average value of the temperature sensor of the ironing equipment is >180℃, the temperature over-limit alarm will be triggered directly.

[0090] Multi-condition joint matching: If the time-domain mean of the sewing machine thread tension is >30N and the variance is >5N², it is determined that the thread tension is abnormal, triggering the "automatic adjustment of thread tension parameters" command.

[0091] (4) Decision implementation and feedback

[0092] Upon successful matching, the rule engine immediately generates decision instructions (such as alarm signals and parameter adjustment values), which are transmitted to the execution unit of the corresponding device (such as the tension controller of a sewing machine or the drive module of a cutting bed) via industrial communication protocols, with an execution latency of ≤10ms. Simultaneously, the decision results are recorded in the local logs of the edge nodes for subsequent collaborative optimization of the rule base with the cloud.

[0093] 2. Decision-making steps of lightweight SVM models

[0094] The lightweight SVM (Support Vector Machine) model uses classification learning to determine device status, making it suitable for decision-making scenarios requiring complex feature fusion (such as determining cropping accuracy and identifying hidden device faults). The specific steps are as follows:

[0095] (1) Model training and lightweight deployment

[0096] Training data preparation: Preprocessed data collected in historical production and corresponding labels (such as "cutting qualified / unqualified" and "equipment normal / minor fault") are used to extract time domain (mean, peak value), frequency domain (main frequency, spectral energy), and time-frequency (wavelet transform coefficients) features as input vectors.

[0097] Model training: Construct an SVM classifier using the radial basis function (RBF) kernel function and optimize the kernel parameters (e.g., γ=0.1) to achieve a classification accuracy of ≥95% on the validation set; compress the model size to ≤5MB by pruning redundant support vectors to adapt to edge node hardware resources (e.g., NVIDIA Jetson series development boards).

[0098] Deployment: The trained lightweight SVM model is embedded into the local decision unit of the edge computing node, supporting real-time inference calls.

[0099] (2) Real-time feature input and preprocessing

[0100] Receive the real-time feature data extracted in step S2 (such as the time-domain variance of cutting bed pressure and the frequency domain energy value of 10-50Hz), sort them according to the feature dimensions during training, normalize them to the model adaptation range (such as the [0,1] interval), and form the input feature vector.

[0101] (3) Model reasoning and classification decision

[0102] Feature mapping: The SVM model maps the input feature vector to a high-dimensional space through a kernel function and calculates the distance to the support vectors of each class.

[0103] Classification and Judgment: Based on the distance determination result, output classification labels (such as "Cutting Accuracy Grade A" or "Slight Thread Jam in Sewing Machine"), with an inference delay of ≤8ms. For example, when the model outputs the label "Tool Wear" based on the time-frequency characteristics of the cutting bed vibration signal, it determines that the machine needs to be stopped for inspection.

[0104] (4) Implementation and feedback of decision results

[0105] Edge nodes convert the classification results of the SVM model into specific execution instructions (such as "cutting bed tool replacement prompt" and "sewing machine operation parameter fine-tuning") and transmit them to the control unit of the corresponding device. At the same time, they cache the inference results and original feature data for subsequent incremental model training (and collaboratively update the model with the cloud).

[0106] Cloud Management Layer

[0107] A deep analytics platform built on the TensorFlow framework:

[0108] Data storage cluster: Supports distributed storage of petabyte-scale production data.

[0109] Intelligent analysis engine: integrates a CNN-LSTM hybrid model (equipment failure prediction accuracy ≥90%).

[0110] Decision feedback system: Enables production strategies to be pushed out in seconds.

[0111] Data processing method flow:

[0112] Multi-source data acquisition and preprocessing:

[0113] Differential sampling: Sewing machines use 500ms high-frequency sampling, while ironing equipment uses 5s low-frequency sampling.

[0114] Level 3 data cleaning:

[0115] ① Threshold filtering: Data with sewing machine speeds exceeding the range of 2000-4000 rpm are automatically marked.

[0116] ② Kalman filtering: After noise suppression, the fluctuation range of temperature data is ≤ ±0.5℃.

[0117] ③ Normalization: Line tension (0-50N) and rotation speed (0-6000 rpm) are uniformly mapped to the [0,1] range.

[0118] Feature fusion and local decision making

[0119] Three-domain feature extraction:

[0120] Time domain: Calculate the 30-second moving average and variance of the line tension data.

[0121] Frequency domain: Perform a 1024-point FFT transform on the cutting bed vibration signal to extract the energy in the 10-50Hz frequency band.

[0122] Time and frequency: The sewing machine current signal was decomposed into three levels using the db4 wavelet.

[0123] Two-tier decision-making mechanism:

[0124] Rule-based decision: When the mean line tension is greater than 30N and the variance is greater than 5N², the equipment parameters are automatically adjusted.

[0125] Model decision: The pruning accuracy is determined by using a trained decision tree model (number of nodes ≤ 128).

[0126] Hierarchical Collaboration Mechanism

[0127] Adaptive transmission strategy:

[0128] Incremental transmission: Rotation speed data changes by less than 5% without being transmitted again.

[0129] Scheduled transmission: Accumulated data is transmitted in packets every hour (compression ratio ≥3:1).

[0130] In-depth cloud analytics:

[0131] Predict the remaining lifespan of equipment based on an RNN model (error ≤ 5%).

[0132] Identify fabric defects using CNN (accuracy ≥ 92%).

[0133] Optimize production scheduling (increase capacity by 15%).

[0134] Implementation Case:

[0135] I. Hardware Deployment and Implementation

[0136] 1. Edge node configuration

[0137] Hardware selection: NVIDIA Jetson AGX Orin development board, configured with 64GB RAM / 512GB SSD, integrated with 2.5G Ethernet interface and 5G module, supporting real-time data processing and dual-link communication.

[0138] Deployment method: Installed in the equipment control cabinet via DIN rail, with an operating temperature range of -10℃ to 60℃, meeting the industrial environment requirements of garment production lines.

[0139] 2. Sensor Network Construction

[0140] Cutting unit: The cutting table surface is embedded with an 8×8 pressure sensor array (spatial resolution 1cm×1cm, measuring range 0-500N) and a position sensor (positioning accuracy 0.5mm).

[0141] Sewing unit: Each sewing machine is equipped with a 3-axis accelerometer (measuring range ±16g), a non-contact speed sensor (resolution 1 rpm), a sampling frequency of 100Hz, and a thread tension sensor (range 0-50N, accuracy 0.1N).

[0142] Communication protocol: The sensor adopts the Modbus RTU protocol with a data transmission baud rate of 115200bps to ensure real-time performance.

[0143] II. Software Implementation Details

[0144] 1. Data Preprocessing Module

[0145] (1) Kalman filter noise reduction algorithm

[0146] For environmental noise suppression of temperature sensors (such as those with a 0-300℃ temperature range in ironing equipment), the discrete Kalman filter iterative formula is adopted:

[0147] Prediction steps:

[0148] Update steps:

[0149]

[0150] Parameter description:

[0151] : The predicted state value at time k (such as the predicted value of temperature and rotation speed).

[0152] : State transition matrix, describing the relationship between the state of the device from time k-1 to time k (such as the dynamic model parameters of the sewing machine speed).

[0153] : Control input item, B k For the control matrix, u k For control quantities (such as equipment preset parameters).

[0154] Prediction error covariance reflects the uncertainty of the predicted value.

[0155] Process noise covariance describes the noise in the dynamic model of the system (such as mechanical vibration interference during equipment operation).

[0156] Kalman gain is used to balance the weights of predicted and measured values.

[0157] : Observation matrix, which maps the state space to the observation space (such as the measurement characteristic matrix of a sensor).

[0158] : Observation noise covariance, which describes the measurement noise of the sensor (such as environmental interference of the temperature sensor).

[0159] : The sensor's measured value at time k (e.g., the real-time temperature collected by a temperature sensor).

[0160] : Identity matrix.

[0161] The measured noise standard deviation was reduced by 85%, and the temperature data fluctuation range was controlled within ±0.5℃.

[0162] (2) Minimum-Maximum Normalization

[0163] Normalize the sewing machine speed v (0-6000 rpm) and thread tension T (0-50 N):

[0164] Parameter description:

[0165] Raw data (such as measured values ​​of sewing machine speed and thread tension).

[0166] The minimum value of a data sequence (e.g., the historical minimum value of rotational speed).

[0167] : The maximum value of the data sequence (e.g., the historical maximum speed).

[0168] The normalized data is mapped to the interval [-1, 1].

[0169] Heterogeneous data are uniformly mapped to the [-1,1] interval to solve the problem of dimensional differences and facilitate subsequent feature extraction.

[0170] 2. Feature Extraction Module

[0171] (1) Calculation of time-domain features

[0172] 30-second moving average and variance of line tension data:

[0173] Where n is the number of sampling points (e.g., 30 seconds × 100Hz = 3000 points). Reflecting average tension, It reflects the degree of fluctuation.

[0174] Parameter description:

[0175] : Preprocessed data sequence (such as real-time sampled values ​​of line tension).

[0176] Number of sampling points (e.g., the number of samplings within 30 seconds).

[0177] The mean of the data reflects the average level of the data (such as the average tension of a line).

[0178] Data variance reflects the degree of data fluctuation (such as the stability of line tension).

[0179] (2) Frequency domain feature extraction

[0180] Perform a 1024-point FFT transform on the cutting bed vibration signal:

[0181] Extracting energy from the 10-50Hz frequency band:

[0182] Used to identify fault frequencies such as tool wear (e.g., the main frequency is about 20Hz during normal cutting).

[0183] Parameter description:

[0184] Time-domain signals (such as time series of cutting bed vibration signals).

[0185] Number of sampling points (e.g., 1024-point FFT transform).

[0186] : Frequency index, corresponding to different frequency components.

[0187] Frequency domain amplitude, reflecting the intensity of each frequency component.

[0188] : Spectral energy, describes the energy distribution of a signal in the frequency domain (used to identify equipment fault frequencies).

[0189] (3) Time-frequency feature extraction

[0190] The sewing machine current signal was decomposed into three levels using the db4 wavelet:

[0191]

[0192] Parameter description:

[0193] Basic wavelet functions (such as db4 wavelet).

[0194] Scale parameter, which controls the scaling of the wavelet function (e.g., 3-level decomposition corresponds to different frequency resolutions).

[0195] Translation parameter: controls the translation position of the wavelet function.

[0196] : Discretized wavelet basis functions.

[0197] The signal to be analyzed (such as the sewing machine current signal).

[0198] Wavelet transform coefficients reflect the characteristics of the signal at different scales and locations.

[0199] in =1, 2, 3 are scale parameters, which are used to obtain time-frequency features at different frequency resolutions and identify abnormal device start-up and shutdown.

[0200] 3. Local Decision Module

[0201] (1) Rule engine decision

[0202] When the average tension And variance When triggered, parameter adjustment:

[0203] (2) Lightweight SVM model

[0204] Decision function based on cutting position x and pressure p:

[0205] Radial basis kernel function:

[0206] Parameter description:

[0207] Input features (such as the location of the cutting bed and the measured pressure value).

[0208] : Feature values ​​of training samples (such as historical cropping locations and stress data).

[0209] Number of training samples.

[0210] Lagrange multipliers are determined by the training process.

[0211] : Labels of the training samples (e.g., “cropping accurate” or “cropping abnormal”).

[0212] Decision threshold, used for classification and judgment.

[0213] : Kernel function parameters, which control the complexity of the classification boundary.

[0214] Euclidean distance measures the similarity of feature vectors.

[0215] in The kernel parameters are defined as follows: model size ≤ 5MB, inference latency ≤ 8ms, and classification accuracy ≥ 95%.

[0216] 4. Edge-Cloud Collaboration Module

[0217] (1) Incremental transmission strategy

[0218] Sewing machine speed increment transmission trigger condition:

[0219] Parameter description:

[0220] : The measured value of the rotational speed at time k.

[0221] : The measured value of the rotational speed at time k-1.

[0222] : Relative change threshold (value range 0.01-0.05, configurable).

[0223] Reference speed value (e.g., the rated speed of a sewing machine)

[0224] Data is transmitted only when the speed change exceeds 5% of the reference value; otherwise, it is not transmitted repeatedly, resulting in a 78% reduction in the measured data transmission volume.

[0225] (2) Cloud-based LSTM fault prediction model

[0226] Equipment operation data sequence Modeling:

[0227] Parameter description:

[0228] : Input data at time t (such as a sequence of device operating parameters).

[0229] : The hidden state at time t-1.

[0230] : The cell state at time t-1.

[0231] Forget gate output controls the degree of forgetting of cell states.

[0232] Input gate output controls the input of new information.

[0233] : Candidate cell state, generated from input data.

[0234] : The cell state at time t, storing long-term memory.

[0235] Output gate output, controls the output in the hidden state.

[0236] : The hidden state at time t, which includes short-term memory.

[0237] : Weight matrix, learned through training.

[0238] : Bias vector, learned through training.

[0239] Activation function, output range [0, 1].

[0240] tanh: Hyperbolic tangent activation function, output range [-1, 1].

[0241] Element-wise multiplication.

[0242] In summary, the present invention has the following beneficial effects:

[0243] Improved data processing efficiency: Edge devices process data in real time, reducing the amount of data transmitted to the cloud, lowering network latency, and improving the speed and real-time performance of data processing.

[0244] Enhanced system reliability: The edge computing layer offloads some of the computing tasks from the cloud, reducing reliance on the central server and improving system reliability and stability.

[0245] Supports intelligent decision-making: Local decision-making algorithms can respond promptly to anomalies in the production process, while cloud-based in-depth analysis provides a basis for optimizing the production process, supporting the intelligent operation of flexible garment production lines.

[0246] The technical solutions provided by the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The descriptions of the embodiments above are only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An edge computing system for data acquisition of intelligent body equipment in a flexible garment production line, characterized in that, include: The intelligent agent device layer includes various devices and supporting sensors on the flexible garment production line, used to collect equipment operation data; The edge computing layer, deployed in the area surrounding the intelligent agent device layer, consists of edge computing nodes used to receive and process data and make local decisions. The cloud management layer is used to perform in-depth analysis, storage, and global decision-making on the data uploaded from the edge computing layer, and then feed it back to the edge computing layer. After the sensors in the intelligent agent device layer collect device operation data according to the set frequency and format, they transmit the data in real time to the edge computing nodes in the edge computing layer through the industrial communication protocol. After receiving the data and completing preprocessing, feature extraction and local decision-making, the edge computing nodes generate device control commands and transmit them back to the corresponding devices in the intelligent agent device layer through the original communication protocol to realize real-time control of the devices.

2. The edge computing system for data acquisition of intelligent body equipment in flexible garment production lines according to claim 1, characterized in that: The sensors in the intelligent device layer include: a speed sensor, a thread tension sensor, and a stitch distance sensor installed on a sewing machine; a position sensor and a pressure sensor installed on a cutting bed; and a temperature sensor and a humidity sensor installed on an ironing device.

3. The edge computing system for data acquisition of intelligent body equipment in flexible garment production lines according to claim 1, characterized in that: The edge computing node uses NVIDIA Jetson series development boards as its hardware platform.

4. An edge computing method for data acquisition of intelligent body equipment in a flexible garment production line, applied to the system described in any one of claims 1-3, characterized in that, Includes the following steps: Step S1, Data Acquisition and Preprocessing: Sensors in the intelligent agent device layer acquire device operation data according to a set frequency and format, and transmit it to the edge computing nodes in the edge computing layer through an industrial communication protocol; the edge computing nodes clean, reduce noise, and normalize the received raw data to generate preprocessed data; Step S2, Feature Extraction and Local Decision-Making: Edge computing nodes extract features from the preprocessed data to obtain time-domain features, frequency-domain features, and time-frequency features; based on the extracted features, local decisions are made through a rule engine or a lightweight machine learning model to generate device control commands or data to be transmitted. Step S3, Edge-Cloud Collaboration: Edge computing nodes transmit data requiring in-depth analysis in local decision-making to the cloud management layer using incremental and timed transmission strategies. The cloud management layer performs in-depth analysis on the received data and generates global decision results, which are then fed back to the edge computing nodes via the network. After receiving the feedback results, the edge computing nodes combine them with their local decision-making logic to execute subsequent operations, forming a closed-loop data flow.

5. The edge computing method for data acquisition of intelligent body equipment in flexible garment production lines according to claim 4, characterized in that: In the data acquisition and preprocessing process, outliers are removed by setting a threshold, and the Kalman filter algorithm is used to reduce noise in the data, normalizing data of different dimensions to the [-1,1] interval.

6. The edge computing method for data acquisition of intelligent body equipment in flexible garment production lines according to claim 4, characterized in that: The feature extraction includes: Temporal feature extraction involves performing sliding window calculations on preprocessed continuous time-series data to obtain the mean, variance, and peak value of the data, reflecting the overall trend and fluctuation characteristics of the data in the time dimension. Frequency domain feature extraction involves converting the same set of data after time domain analysis to the frequency domain using Fourier transform to obtain the dominant frequency and spectral energy, which are used to identify the periodic features hidden in the data. Time-frequency feature extraction is used for complex data that contains both time-domain trends and frequency-domain characteristics. Based on the analysis of time-domain and frequency-domain features, wavelet transform is used to perform joint time-frequency domain analysis. At the same time, the variation pattern of the data on the time axis and the distribution characteristics of different frequency components are preserved, so as to achieve multi-dimensional feature fusion of the preprocessed data.

7. The edge computing method for data acquisition of intelligent body equipment in flexible garment production lines according to claim 4, characterized in that: In step S2, the extracted feature data is correlated and judged based on preset rules. The time domain mean, variance and frequency domain main frequency of the sewing machine thread tension are combined. When the time domain mean exceeds the upper limit and the variance is large, it is determined that the thread tension is abnormal and an alarm is triggered. Alternatively, a learning model such as a decision tree or support vector machine can be deployed, using time-domain features as the basic input features and frequency-domain features and time-frequency features as supplementary features.

8. The edge computing method for data acquisition of intelligent body equipment in flexible garment production lines according to claim 4, characterized in that: The incremental transmission strategy is as follows: when the device data does not change within a set time, it will not be transmitted repeatedly, but only the changed or newly added data will be transmitted.

9. The edge computing method for data acquisition of intelligent body equipment in flexible garment production lines according to claim 4, characterized in that: The timed transmission strategy is to batch-package and transmit the data accumulated by the edge devices to the cloud at preset time intervals.

10. The edge computing method for data acquisition of intelligent body equipment in a flexible garment production line according to claim 4, characterized in that: The cloud-based deep data analysis employs deep learning algorithms such as convolutional neural networks or recurrent neural networks to model and predict the data, and then feeds the results back to the edge computing layer.

Images