Method and system for monitoring operation state of reflow soldering equipment

Abstract

The invention relates to the technical field of data processing, and discloses a method and system for monitoring the running state of reflow soldering equipment. The method comprises the steps of collecting data through a multi-point sensor, extracting temperature gradient, tension fluctuation and gas concentration characteristics, constructing a state recognition model, calculating a health index and setting an early warning threshold value, constructing a fault precursor extraction model to predict a future state, and finally establishing a maintenance strategy optimization system and generating an equipment maintenance plan based on fault early warning information. Through multi-dimensional data fusion and intelligent analysis, early accurate identification of the abnormal state of the equipment, accurate prediction of the fault development trend and active maintenance decision based on quality influence and cost optimization are realized, so that the welding quality stability is improved, the non-planned downtime is shortened, and the maintenance cost is reduced.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method and system for monitoring the operating state of a reflow soldering device. Background Art

[0002] Reflow soldering is a key process used in surface mount technology (SMT) in modern electronics manufacturing, and is widely applied to the printed circuit board (PCB) assembly production line. The monitoring of the operating state of traditional reflow soldering devices mainly relies on a single-point alarm mechanism triggered by preset thresholds, and operators judge the device state based on experience and perform maintenance. With the development of electronic products towards high density and miniaturization, the requirements for soldering accuracy and quality are constantly increasing. Some monitoring systems have begun to adopt simple data acquisition and statistical analysis methods, such as the mean-standard deviation discrimination method, trend graph analysis, etc., to monitor the key parameters of the device in real time. Some advanced systems have introduced rule-based expert systems to judge abnormal states through predefined if-then rules and perform equipment maintenance in combination with a regular maintenance plan.

[0003] However, the existing monitoring technologies for reflow soldering devices have obvious deficiencies. First, single-point parameter monitoring cannot comprehensively reflect the collaborative working state of the complex thermodynamic, mechanical, and gas systems of the device, resulting in inaccurate or delayed abnormal detection; second, the method of presetting fixed thresholds is difficult to adapt to different production processes and environmental conditions, causing false alarms or missed alarms; third, simple statistical analysis methods cannot identify the complex correlation relationships and fault evolution laws between parameters and lack the ability to predict the development trend of faults; fourth, the existing maintenance strategies are mostly passive responses or fixed cycles, and fail to comprehensively optimize the device state, production plan, and maintenance cost, resulting in unnecessary downtime losses or premature / late maintenance interventions; finally, the maintenance decision-making lacks a direct association with product quality and it is difficult to achieve a balance between quality assurance and cost control. Summary of the Invention

[0004] This application provides a method and system for monitoring the operating state of a reflow soldering device, which is used to achieve early and accurate identification of abnormal device states, accurate prediction of the development trend of faults, and proactive maintenance decision-making based on quality impact and cost optimization through multi-dimensional data fusion and intelligent analysis, thereby improving the stability of soldering quality, reducing unplanned downtime, and reducing maintenance costs.

[0005] In a first aspect, the present application provides a method for monitoring the operating state of a reflow soldering device. The method for monitoring the operating state of the reflow soldering device includes: collecting operating parameters through multi-point sensors of the reflow soldering device to obtain a soldering device state data set; extracting a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the soldering device state data set, and performing multi-dimensional feature cross-correlation analysis to obtain a soldering device operating feature matrix; constructing a state recognition model based on the soldering device operating feature matrix, classifying the device operating state through a dynamic threshold adaptive adjustment mechanism to obtain a soldering device state category; calculating a device health index according to the soldering device state category, setting multi-level warning thresholds to obtain a soldering device health assessment result; constructing a fault precursor feature extraction and amplification model using the soldering device health assessment result, predicting the future state through comparison with a reflow soldering feature time series pattern library to obtain soldering device fault warning information; constructing a multi-objective maintenance strategy optimization system based on the soldering device fault warning information, generating a maintenance plan through correlation analysis of welding quality and device state to obtain a soldering device maintenance plan.

[0006] In a second aspect, the present application provides a system for monitoring the operating state of a reflow soldering device. The system for monitoring the operating state of the reflow soldering device includes: An acquisition module, configured to collect operating parameters through multi-point sensors of the reflow soldering device to obtain a soldering device state data set; An association module, configured to extract a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the soldering device state data set, and perform multi-dimensional feature cross-correlation analysis to obtain a soldering device operating feature matrix; A construction module, configured to construct a state recognition model based on the soldering device operating feature matrix, classify the device operating state through a dynamic threshold adaptive adjustment mechanism to obtain a soldering device state category; A calculation module, configured to calculate a device health index according to the soldering device state category, set multi-level warning thresholds to obtain a soldering device health assessment result; A comparison module, configured to construct a fault precursor feature extraction and amplification model using the soldering device health assessment result, predict the future state through comparison with a reflow soldering feature time series pattern library to obtain soldering device fault warning information; A generation module, configured to construct a multi-objective maintenance strategy optimization system based on the soldering device fault warning information, generate a maintenance plan through correlation analysis of welding quality and device state to obtain a soldering device maintenance plan.

[0007] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-mentioned method for monitoring the operating state of a reflow soldering device.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is caused to execute the above-mentioned method for monitoring the operating state of a reflow soldering device.

[0009] In the technical solution provided by this application, a comprehensive welding equipment status dataset is obtained by collecting operating parameters through multi-point sensors, realizing the comprehensive monitoring of multiple systems and parameters of the equipment. Compared with the traditional single-point monitoring method, the comprehensiveness and accuracy of data collection are greatly improved, providing a rich data basis for subsequent analysis. The temperature gradient distribution map, conveyor belt tension fluctuation characteristics, and gas concentration time-varying curve are extracted, and multi-dimensional feature cross-correlation analysis is carried out to obtain the welding equipment operation feature matrix. This process fully explores the internal correlation between multi-dimensional parameters, reveals the mutual influence mechanism between different subsystems, and enables the detection of abnormal states not to be limited to the judgment of single-parameter over-limitation, but to be evaluated from the overall performance of the system. The state recognition model constructed based on the welding equipment operation feature matrix realizes the accurate classification of the equipment operation state through a dynamic threshold adaptive adjustment mechanism. This mechanism dynamically adjusts the classification threshold according to the state transition frequency and duration in the equipment operation historical data, effectively solving the misjudgment problem caused by fixed thresholds and improving the accuracy and stability of state recognition. Especially in the combined application of three machine learning algorithms, namely support vector machine, random forest, and gradient boosting decision tree, the respective advantages of different algorithms are fully utilized. The support vector machine is good at dealing with high-dimensional feature spaces and complex decision boundaries, the random forest has good anti-noise ability and feature importance evaluation ability, and the gradient boosting decision tree is good at dealing with unbalanced data and capturing weak feature changes. The weighted fusion of the three algorithms significantly improves the classification performance. The equipment health index is calculated according to the welding equipment status category, and multi-level warning thresholds are set, realizing the transformation from state classification to health quantification, making the equipment state evaluation more refined and intuitive. A fault precursor feature extraction and amplification model is constructed using the welding equipment health assessment results, and the future state is predicted by comparing with the reflow welding feature time series pattern library, transforming the passive response into an active prediction, greatly advancing the fault warning time and gaining sufficient preparation time for maintenance decisions. A multi-objective maintenance strategy optimization system is constructed based on the welding equipment fault warning information. A maintenance plan is generated through the correlation analysis of welding quality and equipment status, introducing the product quality factor into the maintenance decision-making process for the first time and realizing the transformation from a simple equipment orientation to the coordinated optimization of quality and cost. This system solves for the optimal maintenance time window through an integer programming algorithm, minimizing the total maintenance cost while ensuring welding quality, providing scientific and economic maintenance decision support for enterprises, significantly improving production efficiency and product yield, and reducing maintenance costs and unplanned downtime losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1 It is a schematic diagram of an embodiment of the method for monitoring the operating state of a reflow soldering device in an embodiment of the present application; Figure 2 It is a schematic diagram of an embodiment of the system for monitoring the operating state of a reflow soldering device in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. Detailed implementation manners

[0012] The embodiments of the present application provide a method and a system for monitoring the operating state of a reflow soldering device. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific processes of the embodiments of the present application are described below. Please refer to Figure 1 An embodiment of the method for monitoring the operating state of a reflow soldering device in an embodiment of the present application includes: Step S101: Collect operating parameters through multi-point sensors of the reflow soldering device to obtain a soldering device state data set; Step S102: Extract a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the soldering device state data set, and perform multi-dimensional feature cross-correlation analysis to obtain a soldering device operation feature matrix; Step S103: Construct a state recognition model based on the soldering device operation feature matrix, classify the device operating state through a dynamic threshold adaptive adjustment mechanism, and obtain the soldering device state category; Step S104: Calculate the device health index according to the soldering device state category, and set multi-level warning thresholds to obtain the soldering device health assessment result; Step S105: Use the soldering device health assessment result to construct a fault precursor feature extraction and amplification model, and predict the future state by comparing with the reflow soldering feature time series pattern library to obtain the soldering device fault warning information; Step S106: Construct a multi-objective maintenance strategy optimization system based on the welding equipment failure warning information, generate a maintenance plan through the correlation analysis of the welding quality and the equipment status, and obtain the welding equipment maintenance plan.

[0014] It can be understood that the execution subject of this application can be the operation status monitoring system of the reflow welding equipment, or a terminal or a server. Specifically, it is not limited here. In the embodiments of this application, the server is taken as the execution subject for illustration.

[0015] Specifically, the implementation of multi-point sensors for collecting operating parameters first installs temperature sensor arrays in the preheating zone, reflow zone, and cooling zone of the reflow soldering equipment. These temperature sensor arrays are grid-like structures composed of multiple PT100 platinum resistance temperature sensors arranged longitudinally and horizontally along the soldering furnace body. A speed sensor and a tension sensor are installed in the conveyor belt system. The speed sensor uses an optical encoder, and the tension sensor uses a strain gauge force sensor. A gas concentration sensor and a gas pressure sensor are installed in the gas supply system. The gas concentration sensor is used to monitor the concentration of nitrogen or protective gas, and the gas pressure sensor is used to monitor the gas delivery pressure. A current-voltage sensor is installed in the power supply system to monitor the power parameters of each heating zone. All sensors are uniformly sampled through an industrial Internet of Things module. The sampling frequency of temperature data is 10 Hz, the sampling frequency of conveyor belt parameters is 5 Hz, the sampling frequency of gas parameters is 2 Hz, and the sampling frequency of electrical parameters is 20 Hz. The collected raw data needs to be preprocessed. The sliding average filtering algorithm is used to denoise the temperature data. This algorithm replaces the value of the current point by taking the average of consecutive multiple sampling points, thereby smoothing out short-term fluctuations. The median filtering algorithm is used for the conveyor belt speed data. This algorithm eliminates occasional interference by taking the median value of the data within a time window. The Kalman filtering algorithm is used for the gas concentration data. This algorithm eliminates the influence of noise and improves signal stability through a prediction-correction iterative process. When extracting features and performing multi-dimensional feature cross-correlation analysis, first segment the preprocessed data according to time windows. The window length is adaptively adjusted to 5 minutes to 30 minutes according to the equipment operation cycle. Calculate the temperature gradient distribution map for the temperature data within each time window. The spatial temperature gradient is obtained by dividing the difference between adjacent temperature sensors by the sensor spacing, forming a temperature gradient spatio-temporal distribution matrix. The temperature gradient distribution map reflects the uniformity and stability of heat energy transfer during the soldering process and is crucial for soldering quality. Extract the tension fluctuation characteristics from the conveyor belt parameters, including the peak-to-peak value of tension, the tension stability coefficient, and the tension pulsation frequency characteristics. The peak-to-peak value of tension refers to the difference between the maximum and minimum values of tension. The tension stability coefficient is the ratio of the standard deviation of tension to the average value. The tension pulsation frequency characteristics are the main frequency components obtained by performing a fast Fourier transform on the tension time series data. Perform time-frequency transformation on the gas concentration data and extract the time-varying curve characteristics of gas concentration, including the gas concentration change rate, the gas concentration fluctuation period, and the proportion of the gas concentration stable interval. These characteristics are dimensionally reduced by the principal component analysis method, and the principal components with a cumulative contribution rate reaching 95% are retained to form the operating characteristic matrix of the soldering equipment.

[0016] The construction of the status recognition model first pre - defines five typical operating states of the reflow soldering equipment: normal operating state, temperature anomaly state, conveyor belt anomaly state, gas system anomaly state, and electrical system anomaly state. The operation feature matrix of the soldering equipment is divided into a training matrix and a verification matrix, with the ratio usually being 70% and 30%. Three different classifiers are constructed using the support vector machine algorithm, random forest algorithm, and gradient - boosting decision tree algorithm respectively. The support vector machine algorithm realizes state classification by finding the optimal classification hyperplane. The random forest algorithm determines the state type through the voting results of multiple decision trees. The gradient - boosting decision tree algorithm improves the classification performance by iteratively training a series of weak classifiers and combining them with weights. A dynamic threshold adaptive adjustment mechanism is constructed to dynamically adjust the discriminant threshold parameters of the three classifiers according to the state transition frequency and duration in the historical operation data of the equipment, improving the classification accuracy. Finally, the discriminant results of the three classifiers are fused by the weighted voting method to obtain the final status category of the soldering equipment. The health index calculation method first converts the status category of the soldering equipment into the subsystem status quantization value. The normal operating state is assigned a value of 1.0, the mild anomaly state is assigned a value of 0.8, the moderate anomaly state is assigned a value of 0.6, and the severe anomaly state is assigned a value of 0.3. According to the influence degree of different subsystems on the overall performance of the equipment, subsystem weight coefficients are assigned. Usually, the weight of the temperature system is 0.4, the weight of the conveyor belt system is 0.3, the weight of the gas system is 0.2, and the weight of the electrical system is 0.1. Multiply the subsystem status quantization value by the subsystem weight coefficient to obtain the weighted status score. Extract the anomaly duration feature from the historical changes of the soldering equipment status category to obtain the anomaly duration factor. Based on the correlation of multi - subsystem anomalies in the soldering equipment status category, quantify the diffusion degree of the anomaly within the system to obtain the anomaly diffusion factor. Combine the weighted status score, anomaly duration factor, and anomaly diffusion factor to calculate the final equipment health index, and set multi - level warning thresholds. Usually, a health index lower than 0.9 is a mild warning, lower than 0.7 is a moderate warning, and lower than 0.5 is a severe warning.

[0017] Fault precursor feature extraction and future state prediction: First, associate the health assessment results of the welding equipment with the equipment parameters at historical time points to construct a time-series mapping database. Extract the fault evolution patterns from the time-series mapping database, identify the characteristic change rules before the decline of the equipment health index, and obtain the fault precursor feature set. Perform non-linear amplification processing on the weak change signals in the fault precursor feature set, and usually set the amplification factor to 2 to 5 times to make the small trend changes easier to identify. Collect the typical evolution paths and characteristic sequences of different types of faults to form a characteristic time-series pattern library for reflow soldering. Calculate the similarity between the amplified fault symptom features and the characteristic time-series pattern library for reflow soldering, and use the Euclidean distance or DTW (Dynamic Time Warping) algorithm to calculate the similarity of the characteristic sequences to identify potential fault types and development trends. Based on the fault matching degree score and the change trend of the equipment health index, predict the future equipment state and the possible fault risks. Maintenance strategy optimization system construction: First, map and associate the fault warning information of the welding equipment with the equipment physical structure and operation logic, analyze the correlation between the historical welding quality data and the equipment state parameters, and obtain the quality impact factor matrix. Calculate the equipment downtime loss, maintenance labor cost, and spare parts cost for different maintenance strategies to obtain the maintenance cost estimation table. Use the quality impact factor matrix and the maintenance cost estimation table as constraints, and calculate the optimal maintenance time window through the integer programming algorithm. According to the fault type in the fault warning information and the equipment structure information, generate specific maintenance items and operation procedures to form a complete maintenance plan including maintenance time, maintenance items, required personnel, and resource allocation.

[0018] Taking an actual monitoring process as an example, after the temperature sensor array of the reflow soldering equipment has been continuously operating for 24 hours, the temperature data in the middle reflow area fluctuates by 3°C. The temperature gradient between adjacent sensors increases from 5°C / cm in the standard state to 7.2°C / cm. At the same time, the tension stability coefficient in the tension fluctuation characteristics of the conveyor belt increases from 0.05 to 0.12, and the gas concentration change rate increases from 0.5% / min to 1.2% / min. Through feature extraction and cross-correlation analysis, the formed operation feature matrix is determined by the state recognition model as an abnormal temperature state, and the equipment health index drops from 0.95 to 0.82, triggering a mild warning. After the fault precursor feature extraction model amplifies and analyzes the temperature fluctuation characteristics, the comparison with the time-series pattern library shows that the similarity with the heating element aging fault mode reaches 85%, predicting that the heating element may have unstable power problems within the next 48 hours. The maintenance strategy optimization system generates a maintenance plan based on this warning information, suggesting replacing the relevant heating element after the end of the day's production and adjusting the temperature control parameters in the reflow area to avoid sudden shutdowns during the production peak period, minimizing the impact on production and ensuring the welding quality.

[0019] In the embodiments of the present application, a comprehensive welding equipment status dataset is obtained by collecting operating parameters through multi-point sensors, realizing the comprehensive monitoring of multiple systems and parameters of the equipment. Compared with the traditional single-point monitoring method, the comprehensiveness and accuracy of data collection are greatly improved, providing a rich data basis for subsequent analysis; extracting the temperature gradient distribution map, the conveyor belt tension fluctuation characteristics, and the gas concentration time-varying curve, and performing multi-dimensional feature cross-correlation analysis to obtain the welding equipment operation feature matrix. This process fully explores the internal correlation between multi-dimensional parameters, reveals the mutual influence mechanism between different subsystems, enabling the abnormal state detection not to be limited to the judgment of a single parameter exceeding the limit, but to be evaluated from the overall performance of the system; the state recognition model constructed based on the welding equipment operation feature matrix realizes the accurate classification of the equipment operation state through a dynamic threshold adaptive adjustment mechanism. This mechanism dynamically adjusts the classification threshold according to the state transition frequency and duration in the equipment operation historical data, effectively solving the misjudgment problem caused by a fixed threshold, and improving the accuracy and stability of state recognition. Especially in the combined application of three machine learning algorithms, namely support vector machine, random forest, and gradient boosting decision tree, the respective advantages of different algorithms are fully utilized. The support vector machine is good at dealing with high-dimensional feature spaces and complex decision boundaries, the random forest has good anti-noise ability and feature importance evaluation ability, and the gradient boosting decision tree is good at dealing with imbalanced data and capturing weak feature changes. The weighted fusion of the three algorithms significantly improves the classification performance; calculating the equipment health index according to the welding equipment status category and setting multi-level warning thresholds realizes the transformation from state classification to health quantification, making the equipment state evaluation more refined and intuitive; using the welding equipment health assessment result to construct a fault precursor feature extraction and amplification model, predicting the future state by comparing with the reflow welding feature time series pattern library, transforming the passive response into an active prediction, greatly advancing the fault warning time, and winning sufficient preparation time for maintenance decision-making; constructing a multi-objective maintenance strategy optimization system based on the welding equipment fault warning information, generating a maintenance plan through the correlation analysis of welding quality and equipment status, and introducing the product quality factor into the maintenance decision-making process for the first time, realizing the transformation from a simple equipment orientation to the coordinated optimization of quality and cost. This system solves for the optimal maintenance time window through an integer programming algorithm, minimizing the total maintenance cost while ensuring the welding quality, providing scientific and economic maintenance decision support for enterprises, significantly improving production efficiency and product yield, and reducing maintenance costs and unplanned downtime losses.

[0020] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Install temperature sensor arrays in the preheating zone, reflow zone, and cooling zone of the reflow welding equipment, install speed sensors and tension sensors in the conveyor belt system, install gas concentration sensors and pressure sensors in the gas supply system, and install current and voltage sensors in the power supply system to form a multi-type sensor network; (2) Use an industrial Internet of Things module to uniformly sample the data collected by multi-type sensor networks, where the sampling frequency of temperature data is 10 Hz, the sampling frequency of conveyor belt parameters is 5 Hz, the sampling frequency of gas parameters is 2 Hz, and the sampling frequency of electrical parameters is 20 Hz; (3) Use a moving average filtering algorithm to denoise the collected temperature data, use a median filtering algorithm to denoise the conveyor belt speed data, and use a Kalman filtering algorithm to process the gas concentration data; (4) Complement the missing data points using linear interpolation or polynomial interpolation according to their continuous missing lengths, and preliminarily screen the outliers using the statistical 3σ principle; (5) Align the processed sensor data of various types according to a unified timestamp to generate a welding equipment status data set.

[0021] Specifically, installing temperature sensor arrays in the preheating zone, reflow zone, and cooling zone of the reflow soldering equipment is the basis of the monitoring method. These temperature sensor arrays are composed of multiple PT100 platinum resistance temperature sensors arranged in a grid pattern. Usually, 16 - 25 sensors are installed in each zone to form a complete temperature field monitoring network. The temperature range in the preheating zone is typically 150 - 180 °C, the temperature range in the reflow zone is 230 - 260 °C, and the temperature in the cooling zone gradually drops from a high temperature to room temperature. A speed sensor and a tension sensor are installed in the conveyor belt system. The speed sensor uses a high-precision photoelectric encoder and can monitor the running speed of the conveyor belt with an accuracy of 0.1 mm / s. The tension sensor uses a strain gauge force sensor with a measurement range of 0 - 50 N to monitor the change in conveyor belt tension. A gas concentration sensor and a pressure sensor are installed in the gas supply system. The gas concentration sensor monitors the concentration of the protective atmosphere (usually nitrogen) with a measurement range of 0 - 100%, and the pressure sensor has a measurement range of 0 - 1 MPa to monitor the pressure stability of the gas supply system. A current and voltage sensor is installed in the power supply system. The current sensor has a measurement range of 0 - 100 A, and the voltage sensor has a measurement range of 0 - 380 V to monitor the power parameters of the heating element. These multi-type sensor networks perform unified data acquisition through an industrial Internet of Things module and use industrial field bus protocols such as ModBus-RTU or Profibus-DP to achieve communication, ensuring stable and reliable data transmission. Due to the different change rates and importance of different parameters, a differentiated sampling frequency is set: the sampling frequency of temperature data is 10 Hz, that is, temperature data is collected every 0.1 second, which can capture the dynamic changes of temperature; the sampling frequency of conveyor belt parameters is 5 Hz, and speed and tension data are collected every 0.2 second; the sampling frequency of gas parameters is 2 Hz, and gas concentration and pressure data are collected every 0.5 second; the sampling frequency of electrical parameters is 20 Hz, and current and voltage data are collected every 0.05 second to monitor the rapid changes in the electrical system. This differentiated sampling strategy ensures both the monitoring accuracy of key parameters and avoids data redundancy.

[0022] The original data collected usually contains noise and interference and needs signal processing. The sliding average filtering algorithm is used to denoise the temperature data. This algorithm replaces the value of the current point by calculating the average of consecutive N points. The size of the filtering window is usually set to 5 - 9 points, which can effectively smooth the temperature fluctuations without overly delaying the response to temperature changes. The median filtering algorithm is used to denoise the conveyor belt speed data. This algorithm eliminates short-term outliers by taking the median value of the data within a certain time window. The window size is usually 7 points, which can effectively eliminate the spike interference in the conveyor belt speed data. The Kalman filtering algorithm is used to process the gas concentration data. This algorithm establishes a state prediction model and a measurement model for the gas concentration change, and performs recursive optimal estimation by combining the process noise covariance and the measurement noise covariance, which can effectively handle the random fluctuations and systematic errors in the gas concentration data. During the data acquisition process, data loss may inevitably occur due to temporary sensor failures or communication interruptions. The missing data needs to be filled. For data with a continuous missing length of less than 3 sampling points, linear interpolation is used for filling, and the missing value is estimated by linearly connecting adjacent valid data points; for data with a continuous missing length between 3 and 10 sampling points, polynomial interpolation methods such as cubic spline interpolation are used for filling, which can maintain the smoothness and continuity of the data. Outliers are initially screened using the statistical 3σ principle, that is, data outside the range of the mean ± 3 times the standard deviation is marked as a suspicious outlier, and it is decided whether to replace it with an interpolation based on the changing trend of the data before and after.

[0023] Since the sampling frequencies of different sensors are different, in order to perform subsequent multi-sensor fusion analysis, various processed sensor data need to be aligned according to a unified time stamp. Using the minimum common sampling period (0.1 second) as the benchmark, hold interpolation is performed on the low-frequency sampling data, that is, the value of the previous sampling point is held between two sampling points to form a data structure with consistent time. The data set after the above processing includes the temperature field distribution, conveyor belt operation parameters, gas system parameters, and electrical system parameters, constituting a data set that comprehensively reflects the operating state of the welding equipment and providing a data basis for subsequent feature extraction and state recognition.

[0024] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Segment the welding equipment state data set according to different time window lengths, and the window length is adaptively adjusted to 5 minutes to 30 minutes according to different equipment operation cycles; (2) Calculate the temperature gradient distribution map for the temperature data within each time window. The spatial temperature gradient is obtained by dividing the difference between adjacent temperature sensors by the sensor spacing, and then combined with the time dimension to form a temperature gradient spatio-temporal distribution matrix; (3) Extract the conveyor belt tension fluctuation characteristics from the conveyor belt parameters, including the peak-to-peak value of the tension, the tension stability coefficient, and the tension pulsation frequency characteristics, and construct a tension fluctuation feature vector; (4) Perform time-frequency transformation on the gas concentration data, and extract the time-varying curve characteristics of the gas concentration, including the gas concentration change rate, the gas concentration fluctuation period, and the proportion of the gas concentration stable interval; (5) Calculate the phase correlation degree between the temperature gradient distribution map and the conveyor belt tension fluctuation characteristics, analyze the time-delay correlation between the gas concentration time-varying curve and the temperature gradient, and construct a parameter cross-influence matrix; (6) Perform dimensionality reduction processing on the extracted features by the principal component analysis method, retain the principal components with a cumulative contribution rate reaching 95%, and form an operating characteristic matrix of the welding equipment.

[0025] Specifically, the processing of the welding equipment state data set starts with the time window segmentation operation, and the continuously collected data is segmented according to different time window lengths. The window length is adaptively adjusted from 5 minutes to 30 minutes according to the operating cycles of different equipment, and this adaptive window length selection is determined based on the working characteristics of the reflow soldering equipment. For a small reflow soldering equipment on a fast PCB assembly line, a production cycle may only take about 5 minutes, so a 5-minute time window is selected; while for a large reflow soldering equipment when processing large-size circuit boards or multi-layer PCBs, a complete soldering cycle may be up to 30 minutes. The time window is selected by the autocorrelation analysis method to determine the optimal window size, that is, calculate the autocorrelation coefficient of the data at different time intervals, and use the time interval when the autocorrelation coefficient drops below 0.3 as a reference value for the effective window size. In practical applications, a sliding window is used for processing, and there is a 50% overlap between adjacent windows to ensure that the state changes at the window boundaries are not missed. Calculating the temperature gradient distribution map for the temperature data within each time window is a key step in understanding the welding heat distribution. The temperature gradient distribution map is obtained by dividing the difference between adjacent temperature sensors by the sensor spacing to get the spatial temperature gradient. For example, if the spacing between two adjacent sensors is 5 cm, the temperature of one sensor is 200 °C, and the other is 215 °C, then the spatial temperature gradient at this point is (215 - 200) / 5 = 3 °C / cm. For the temperature field of the entire welding equipment, a three-dimensional coordinate system (x, y, z) is used, where x and y represent the position coordinates in the horizontal and vertical directions respectively, and z represents the temperature value. The temperature gradients in each direction are obtained by calculating (∂T / ∂x) and (∂T / ∂y). Combining the time dimension, a temperature gradient spatio-temporal distribution matrix is formed, that is, a two-dimensional temperature gradient matrix is obtained at each sampling time point, and a three-dimensional matrix is formed as it changes with time. The mathematical representation of the temperature gradient spatio-temporal distribution matrix is V(x, y, t), where (x, y) is the spatial position and t is the time, and the matrix element value is the temperature gradient at this position at this time.

[0026] The characteristics of tension fluctuation include the peak-to-peak value of tension, the tension stability coefficient, and the tension pulsation frequency characteristics. The calculation method of the peak-to-peak value of tension is to find the maximum tension value and the minimum tension value within a time window, and the difference between the two is the peak-to-peak value, which reflects the extreme range of tension fluctuation. The tension stability coefficient is obtained by dividing the standard deviation of the tension value by the average value, indicating the stability degree of tension fluctuation relative to the average tension. The tension pulsation frequency characteristics are determined by performing a fast Fourier transform (FFT) on the tension time series data to identify the main frequency components and their amplitudes. For example, for tension data with a sampling rate of 5 Hz, a 512-point FFT analysis is carried out to obtain the frequency spectrum distribution in the range of 0 - 2.5 Hz, and the first three frequencies with the largest amplitudes and their corresponding amplitudes are found to form a frequency-amplitude feature vector with a length of 6. These three characteristics constitute the tension fluctuation feature vector, which is used to characterize the dynamic characteristics of the conveyor belt system. Performing time-frequency transformation on the gas concentration data is a necessary step to understand the performance of the protective gas system. First, continuous wavelet transform (CWT) is used to perform time-frequency analysis on the gas concentration time series, and the Morlet wavelet is selected as the mother wavelet function. After transformation, the time-frequency spectrum of the gas concentration is obtained, showing the energy distribution at different time points and different frequencies. Three types of characteristics are extracted from the time-frequency spectrum: the gas concentration change rate is calculated by the first-order difference of the original concentration data, indicating the change amplitude of the concentration per unit time; the gas concentration fluctuation period is determined by analyzing the frequencies corresponding to the energy peaks in the time-frequency spectrum, and the reciprocals of the three frequency components with the largest energy are taken as the main periods; the proportion of the stable interval of the gas concentration is determined by calculating the proportion of time points where the concentration change rate is less than a preset threshold (usually 0.5% / min). These characteristics comprehensively reflect the stability and dynamic response characteristics of the gas system.

[0027] Calculating the phase correlation degree between the temperature gradient distribution map and the conveyor belt tension fluctuation characteristics is an important means to understand the mutual influence between the temperature field and the mechanical system. The phase correlation degree is calculated using the following formula:

[0028] where, represents the phase correlation degree at a time delay of , is the number of sampling points within the time window, is the number of spatial sampling points of the temperature gradient, is the dimension of the tension feature vector, represents the temperature gradient value at position ( ) at time , represents the tension feature vector at time . By calculating the phase correlation degrees at different time delays , the time delay value corresponding to the maximum correlation degree is found, which reflects the time relationship between the temperature change and the conveyor belt tension change.

[0029] Similarly, the cross-correlation function is used to analyze the time-delay correlation between the time-varying curve of gas concentration and the temperature gradient:

[0030] where, represents the time-delay correlation coefficient between the gas concentration and the average temperature gradient, is the time of the gas concentration, is the time of the average temperature gradient, and are the means of the gas concentration and the average temperature gradient respectively, is the number of sampling points, is the time-delay variable. By calculating the correlation coefficients under different values, a parameter cross-influence matrix is constructed. The elements in this matrix represent the maximum correlation coefficient between parameters and their corresponding time-delay values. Performing dimensionality reduction on the extracted features through the principal component analysis method is an effective means of transforming high-dimensional features into low-dimensional principal components. Principal component analysis first standardizes all features so that the mean of each feature is 0 and the variance is 1, then calculates the covariance matrix, performs eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors. The eigenvectors are sorted in descending order according to the corresponding eigenvalue magnitudes, the contribution rate and cumulative contribution rate of each eigenvalue are calculated, and the first k eigenvectors with a cumulative contribution rate reaching 95% are selected as the projection matrix. Through projection transformation, the original high-dimensional features are converted into k-dimensional principal components, forming the operating feature matrix of the welding equipment.

[0031] Taking an actual reflow soldering process as an example, during the production process of the reflow soldering equipment in an electronic assembly factory, when welding a batch of double-sided PCB boards, a 15-minute time window is set for data analysis according to the characteristics of the production cycle. Within a time window, the temperature gradient distribution map is calculated from the data of 10 temperature sensors in the preheating zone, 12 in the reflow zone, and 8 in the cooling zone. It is found that there is an abnormal temperature gradient in the middle of the reflow zone, and the maximum temperature gradient reaches 8 °C / cm, exceeding the normal working state standard of 5 °C / cm. At the same time, the tension characteristics are extracted from the data collected by the conveyor belt system. The peak-to-peak value of the tension is calculated to be 4.2 N, the tension stability coefficient is 0.08, and the main pulsation frequencies obtained by FFT analysis are 0.12 Hz, 0.25 Hz, and 0.43 Hz, with corresponding amplitudes of 0.7 N, 0.5 N, and 0.3 N respectively. After the gas concentration data is subjected to time-frequency transformation, the gas concentration change rate is extracted as 0.8% / min, the main fluctuation periods are 180 s, 90 s, and 45 s, and the stable interval ratio is 78%. The phase correlation analysis of the temperature gradient and the tension fluctuation is carried out, and the maximum correlation coefficient is calculated to be 0.72, with a corresponding time delay of 8 s, indicating that the temperature change leads the tension fluctuation by 8 s. The time-delay correlation analysis of the gas concentration and the temperature gradient shows that the maximum correlation coefficient is 0.65, with a corresponding time delay of 15 s, indicating that the temperature gradient change leads the gas concentration change by 15 s. These analysis results are integrated into the parameter cross-influence matrix, and then processed by principal component analysis, reducing from the original 43-dimensional feature space to an 8-dimensional principal component space. The cumulative contribution rate of the 8 principal components reaches 96.4%, effectively reducing the data dimension and retaining the key information. The formed 8×15-dimensional operation feature matrix of the soldering equipment is used as the input data for subsequent state recognition.

[0032] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Pre-define five typical operating states of the reflow soldering equipment: normal operating state, temperature anomaly state, conveyor belt anomaly state, gas system anomaly state, and electrical system anomaly state; (2) Divide the operation feature matrix of the soldering equipment into a training matrix and a verification matrix, and the training matrix is used for subsequent classification algorithm training; (3) Process the training matrix through the support vector machine algorithm to construct a first classifier, and evaluate the performance of the first classifier with the verification matrix to obtain the first classification accuracy rate; (4) Process the training matrix through the random forest algorithm to construct a second classifier, and evaluate the performance of the second classifier with the verification matrix to obtain the second classification accuracy rate; (5) Process the training matrix through the gradient boosting decision tree algorithm to construct a third classifier, and evaluate the performance of the third classifier with the verification matrix to obtain the third classification accuracy rate; (6) Build a dynamic threshold adaptive adjustment mechanism to dynamically adjust the discrimination threshold parameters of the first classifier, the second classifier, and the third classifier according to the state transition frequency and duration in the historical operation data of the device; (7) Based on the first classification accuracy, the second classification accuracy, and the third classification accuracy, fuse the discrimination results of the three classifiers by the weighted voting method to obtain the state category of the welding device.

[0033] Specifically, five typical operating states of the reflow soldering equipment are predefined as the classification targets. The normal operating state means that all parameters of the equipment fluctuate within the set range, the temperature distribution is uniform, the conveyor belt runs smoothly, the gas concentration is stable, and the electrical parameters are normal; the temperature abnormal state means that the temperature gradient distribution map appears abnormal, such as the temperature in a certain area is too high or too low, the temperature fluctuation exceeds the set threshold, and the temperature gradient is greater than the set safety range; the conveyor belt abnormal state means that the conveyor belt speed is unstable, the tension fluctuation exceeds the safety threshold, or the conveyor belt shakes or slips; the gas system abnormal state means that the protective gas concentration fluctuates greatly, the air pressure is unstable, or the gas supply is insufficient; the electrical system abnormal state means that the current and voltage of the heating element fluctuate abnormally, the power fluctuation exceeds the set range, or the electrical control system responds abnormally. Dividing the operation feature matrix of the welding equipment into a training matrix and a verification matrix is a basic step in building a machine learning model. A typical division ratio is that 70% of the data is used for training and 30% of the data is used for verification, or the K-fold cross-validation method is adopted, where the data is evenly divided into K parts, K-1 parts are selected as training data each time, and the remaining 1 part is used as verification data, and the average result is obtained after K times of training. When dividing the data, the stratified sampling method is used to ensure that the proportions of various states in the training set and the verification set are the same, and to avoid model bias caused by uneven data distribution. For the reflow soldering equipment, usually several months of operation data are collected, including various typical working conditions and abnormal states, to form a complete training and verification data set.

[0034] The support vector machine algorithm is a binary classification model that realizes data classification by finding the optimal classification hyperplane. For the multi-class classification problem of the reflow soldering equipment, the one-vs.-rest strategy is adopted, that is, a binary classifier is constructed for each state, and each classifier judges whether the current state belongs to a specific class. The core of the support vector machine is to find the maximum margin hyperplane. For the case of linearly inseparable data, a kernel function is introduced to map the data into a high-dimensional space. The radial basis function (RBF) kernel is usually used for the state recognition of the reflow soldering equipment, and the kernel parameters are determined by grid search and cross-validation to find the optimal values. The training matrix is processed by the support vector machine algorithm to construct the first classifier, and its performance is evaluated using the validation matrix. Metrics such as accuracy, precision, recall, and F1-score are calculated to comprehensively evaluate the performance of the first classifier and obtain the first classification accuracy. The random forest algorithm is an ensemble learning method that constructs multiple decision trees and takes the majority voting result as the final classification result. Each decision tree in the random forest is trained on a random subset of the original training data and considers a random subset of features at each node split, increasing the diversity between trees and reducing the risk of overfitting. For the state recognition of the reflow soldering equipment, the number of decision trees is usually set to 100 - 500, and the maximum tree depth is set to 10 - 20 levels to balance the model complexity and generalization ability. One advantage of the random forest is that it can provide feature importance scores to help identify the key feature parameters for state classification. The training matrix is processed by the random forest algorithm to construct the second classifier, and various performance metrics are calculated using the validation matrix to obtain the second classification accuracy.

[0035] The gradient boosting decision tree algorithm trains a series of decision trees serially. Each new tree fits the residuals of the previous tree, gradually improving the model performance. For the recognition of the reflow soldering equipment status, the common implementations are XGBoost or LightGBM. These algorithms improve the computational efficiency and model performance by optimizing the second derivative of the objective function and regularization techniques. Typical parameter settings include a learning rate of 0.01 - 0.1, a maximum tree depth of 3 - 7, and the number of trees of 100 - 1000. The training matrix is processed through the gradient boosting decision tree algorithm to construct the third classifier, and its performance is evaluated using the validation matrix. Each performance metric is calculated to obtain the third classification accuracy. Constructing a dynamic threshold adaptive adjustment mechanism is a key step to improve the robustness of the classifier. Based on the state transition frequency and duration in the historical operation data of the equipment, the discriminant threshold parameters of each classifier are dynamically adjusted. For the reflow soldering equipment, the state transition frequency is usually low, the normal state duration is long, and the abnormal state is short but requires a quick response. The dynamic threshold adjustment is based on time series analysis, calculating the average duration and transition probability of different states, constructing a Markov state transition model, and dynamically adjusting the discriminant threshold according to the current state and duration. For example, when the equipment is in the normal state for a long time, the threshold for abnormal judgment is appropriately increased to reduce false alarms; when possible abnormal signs are detected, the threshold is decreased to increase sensitivity and capture abnormalities in advance.

[0036] The weighted voting method is an effective way to fuse the results of multiple classifiers. According to the first classification accuracy, the second classification accuracy, and the third classification accuracy, different weights are assigned to the three classifiers, and the weights are proportional to the classification accuracy. For each sample to be classified, the weighted voting results of each classifier are calculated, and the category with the highest number of votes is selected as the final classification result. When the number of votes of multiple categories is close, confidence analysis is used to calculate the vote difference between the category with the highest number of votes and the category with the second highest number of votes. If the difference is less than the preset threshold, it is marked as an uncertain state, triggering a more detailed analysis process.

[0037] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Convert the soldering equipment status category into a subsystem status quantization value; (2) For the temperature abnormal state, conveyor belt abnormal state, gas system abnormal state, and electrical system abnormal state in the soldering equipment status category, assign subsystem weight coefficients to obtain the weighted parameters of the subsystem status quantization value; (3) Multiply the subsystem status quantization value by the subsystem weight coefficient to obtain the weighted status score; (4) Extract the abnormal duration feature from the historical changes of the soldering equipment status category to obtain the abnormal duration factor; (5) Based on the correlation of abnormalities in multiple subsystems within the welding equipment status category, quantify the degree of abnormality diffusion within the system to obtain an abnormality diffusion factor. (6) Perform a comprehensive calculation on the weighted status score, abnormality duration factor, and abnormality diffusion factor to obtain an equipment health index, set multi-level warning thresholds based on the equipment health index, and generate a welding equipment health assessment result.

[0038] Specifically, perform numerical processing on the status categories output by the status recognition model. Specifically, assign a value of 1.0 to the normal operating status, indicating that the equipment is in a completely healthy state; assign a value of 0.8 to the mild abnormality status, indicating that the equipment has a minor abnormality but does not affect normal production; assign a value of 0.6 to the moderate abnormality status, indicating that the degree of equipment abnormality has deepened and requires attention; assign a value of 0.3 to the severe abnormality status, indicating that the equipment has a serious abnormality and requires immediate intervention. The setting of the status quantization value is based on the impact of the reflow welding equipment status on product quality and equipment safety. The closer the value is to 1.0, the closer the status is to the normal operating state. This quantization method enables the status of different subsystems to be compared and comprehensively evaluated on a unified numerical scale. For the temperature abnormality status, conveyor belt abnormality status, gas system abnormality status, and electrical system abnormality status in the welding equipment status category, different subsystem weight coefficients need to be assigned. The subsystem weight coefficient reflects the impact degree of each subsystem on the overall health status of the reflow welding equipment. The weight assignment is based on two main factors: one is the impact degree of the subsystem on welding quality, and the other is the severity of equipment damage caused by subsystem failures. The temperature system directly affects the solder melting and solidification processes and plays a decisive role in welding quality, usually assigned a higher weight, such as 0.4; the conveyor belt system affects the stability and speed uniformity of the PCB board during the welding process and has an important impact on welding quality, usually assigned the second-highest weight, such as 0.3; the gas system mainly affects the oxidation degree of the welding environment and has a certain impact on welding quality, usually assigned a medium weight, such as 0.2; the electrical system, as the basis for equipment energy supply and control, although important, has a relatively small direct impact on welding quality, usually assigned a lower weight, such as 0.1. The sum of these weight coefficients must be equal to 1 to ensure the standardization of the scoring system.

[0039] Multiply the quantized value of the subsystem state by the subsystem weight coefficient to obtain the weighted state score of each subsystem. The calculation formula for the weighted state score is the quantized value of the subsystem state multiplied by the corresponding subsystem weight coefficient, which reflects the weighted contribution of the state of each subsystem to the overall health state. For example, if the temperature system is in a moderately abnormal state (quantized value is 0.6), the conveyor belt system is in a slightly abnormal state (quantized value is 0.8), the gas system and the electrical system are both in a normal state (quantized value is 1.0), then the weighted state score of the temperature system is 0.6×0.4 = 0.24, the weighted state score of the conveyor belt system is 0.8×0.3 = 0.24, the weighted state score of the gas system is 1.0×0.2 = 0.2, and the weighted state score of the electrical system is 1.0×0.1 = 0.1. The sum of the weighted state scores of each subsystem is 0.24 + 0.24 + 0.2 + 0.1 = 0.78, which is used as the basic health score value. Extracting the abnormal duration feature from the historical changes of the welding equipment state category to obtain the abnormal duration factor is an important part of health assessment. The abnormal duration factor reflects the persistence and evolution trend of the equipment's abnormal state. First, count the cumulative time percentage of the equipment in various abnormal states within a recent period (such as 24 hours), and then calculate the abnormal duration factor according to the duration distribution of the abnormal state. The calculation of the abnormal duration factor considers two key parameters: one is the length of continuous abnormal time, and the other is the repetition frequency of the abnormal state. The longer the continuous time and the higher the repetition frequency, the more stable the abnormal state and the higher the failure risk. The specific calculation method uses exponential weighted average, assigning higher weights to abnormal events within the recent time window and lower weights to abnormal events within the earlier time window, so as to more accurately reflect the severity and development trend of the current abnormality.

[0040] Based on the correlation of multi-subsystem abnormalities in the welding equipment state category, quantify the degree of abnormal diffusion within the system. Obtaining the abnormal diffusion factor is a deep-level indicator for evaluating the equipment health state. The abnormal diffusion factor describes whether the fault has spread from one subsystem to other subsystems, reflecting the systematicness and chain reaction degree of the fault. The calculation of the abnormal diffusion factor is based on the fault propagation network model between subsystems. First, establish an association matrix between subsystems, which represents the influence degree of the fault of one subsystem on other subsystems, and then calculate the abnormal diffusion factor according to the current abnormal states of each subsystem and the association matrix. For example, a temperature abnormality usually causes a change in the conveyor belt tension, and a conveyor belt abnormality affects the residence time of the PCB board in each heating zone, which in turn affects the temperature distribution. When multiple related subsystems show abnormalities at the same time, the abnormal diffusion factor value is higher, indicating that the fault has spread within the system and the health condition is more severe.

[0041] Perform a comprehensive calculation on the weighted state score, abnormal duration factor, and abnormal diffusion factor. The formula for obtaining the equipment health index is as follows:

[0042] Among them, represents the device health index (Health Index), represents the state quantization value (Weight Status Value) of the i-th subsystem, represents the weight coefficient (Weight Factor) of the i-th subsystem, n represents the total number of subsystems, ADF represents the abnormal duration factor (Abnormal Duration Factor), APF represents the abnormal propagation factor (Abnormal Propagation Factor), and and are adjustment parameters. Among them, is the basic health score adjustment coefficient, usually set to 1.0; is the abnormal duration impact coefficient, usually set to 0.2 to 0.5; is the abnormal propagation impact coefficient, usually set to 0.3 to 0.6. In this formula, the term ( ) calculates the weighted state score of each subsystem, is the basic health score, the term ( ) represents the adjustment effect of the abnormal duration factor on the health index, and the term ( ) represents the adjustment effect of the abnormal propagation factor on the health index. The product of the three comprehensively considers the comprehensive impact of the current state, duration, and diffusion degree on the device health. According to the calculated device health index, multi-level warning thresholds are set to generate the welding equipment health assessment result. The multi-level warning thresholds are usually divided into three levels: when the health index is lower than 0.9 but higher than 0.7, it is a mild warning, indicating that the device has a minor abnormality and needs attention; when the health index is lower than 0.7 but higher than 0.5, it is a moderate warning, indicating that the device abnormality is obvious and inspection and maintenance need to be arranged; when the health index is lower than 0.5, it is a severe warning, indicating that the device has a serious abnormality and needs to be shut down for repair immediately. The warning thresholds can be fine-tuned according to the specific device characteristics and production requirements to balance the warning sensitivity and accuracy.

[0043] Taking a reflow soldering equipment for producing high - density PCB boards as an example, the process of calculating the health index and early warning is described. During a certain production process, the state recognition model detected fluctuations in the temperature of the reflow area. The temperature gradient distribution map showed that the temperature gradient in the middle area reached 7.8℃ / cm, exceeding the normal value of 5℃ / cm, and it was determined to be in a moderately abnormal temperature state, with a quantization value of 0.6. At the same time, the change rate of the conveyor belt speed increased, and the tension stability coefficient rose from the normal 0.05 to 0.09, which was determined to be in a slightly abnormal conveyor belt state, with a quantization value of 0.8. The parameters of the gas system and the electrical system were normal, and the quantization values were both 1.0. According to the preset subsystem weights (temperature system 0.4, conveyor belt system 0.3, gas system 0.2, electrical system 0.1), the weighted state scores of each subsystem were calculated: temperature system 0.6×0.4 = 0.24, conveyor belt system 0.8×0.3 = 0.24, gas system 1.0×0.2 = 0.2, electrical system 1.0×0.1 = 0.1, and the basic health score was 0.78. By querying the historical data of the equipment, it was found that the abnormal temperature state occurred 3 times within the past 12 hours, and each duration was about 15 minutes. The abnormal duration factor was calculated to be 0.25. The temperature abnormality and the conveyor belt abnormality occurred simultaneously and had an obvious time - series correlation. The conveyor belt abnormality was usually detected within 5 - 10 minutes after the temperature abnormality occurred, indicating that the abnormality had spread from the temperature system to the conveyor belt system. The abnormal diffusion factor was calculated to be 0.4. Substituting these factors into the health index calculation formula, setting \(\alpha = 1.0\), \(\beta = 0.3\), \(\gamma = 0.4\), the equipment health index was 0.78×(1 - 0.3×0.25)×(1 - 0.4×0.4)=0.67, triggering a moderate early warning. The generated health assessment results included detailed information such as the current health index 0.67, the early warning level "moderate early warning", the abnormal subsystems "temperature system (moderately abnormal), conveyor belt system (slightly abnormal)", and the recommended measures "check the calibration status of the heating elements and temperature sensors in the reflow area, check the conveyor belt tension adjustment mechanism", etc.

[0044] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Correlate the health assessment results of the soldering equipment with the equipment parameters at historical time points to obtain a time - series mapping database; (2) Extract the fault evolution patterns from the time - series mapping database, identify the characteristic change rules before the decline of the equipment health index, and obtain a set of fault precursor characteristics; (3) Perform non - linear amplification processing on the weak change signals in the set of fault precursor characteristics to obtain the amplified fault symptom characteristics; (4) Collect the typical evolution paths and characteristic sequences of different types of faults to form a reflow soldering characteristic time - series pattern library; (5) Calculate the similarity between the amplified fault symptom features and the time-sequence pattern library of reflow soldering features to identify potential fault types and development trends, and obtain a fault matching degree score. (6) Based on the fault matching degree score and the change trend of the equipment health index, predict the equipment status at future time points and the possible fault risks, and generate fault warning information for the soldering equipment.

[0045] Specifically, correlate the welding equipment health assessment results with the equipment parameters at historical time points to generate a time-sequence mapping database. This process aligns and correlates the assessment results such as health index and warning level with the original sensor data, feature matrix, and status classification results at the corresponding time points. For each time node, record multi-dimensional data such as the equipment health index, the quantified values of the states of each subsystem, the temperature gradient distribution map, the conveyor belt tension fluctuation characteristics, and the time-varying curve of gas concentration. The correlation method uses the time window matching technique to establish a mapping relationship between the health assessment results and the equipment parameters within the previous time window (such as 5 - 30 minutes), forming a structured time-sequence database containing timestamps, parameter values, and health status. Extracting the fault evolution pattern from the time-sequence mapping database is the key step in identifying fault precursors. By analyzing the parameter change trends before the health index decline event, identify the characteristic change laws during the transition of the equipment from the normal state to the abnormal state. Specific methods include trend analysis, change point detection, and pattern recognition. Trend analysis uses time series decomposition technology to decompose each parameter sequence into a trend term, a periodic term, and a residual term, and focuses on the change slope of the trend term; change point detection uses the CUSUM (cumulative sum) algorithm or the Page-Hinkley detector to identify the mutation points in the parameter sequence; pattern recognition applies the dynamic time warping (DTW) algorithm to search for segments similar to known fault patterns in the parameter sequence. Through these analyses, extract the characteristic change laws that appear before the health index decline, such as the gradually increasing temperature gradient, the increasing conveyor belt tension fluctuation frequency, and the decreasing proportion of the stable interval of gas concentration, to form a fault precursor feature set.

[0046] Nonlinear amplification processing of weak change signals in the fault precursor feature set is an important means to improve the sensitivity of early fault detection. Since the signal changes in the initial stage of a fault are usually very weak and difficult to directly identify, signal amplification techniques are needed to enhance these tiny changes. Nonlinear amplification processing methods include exponential amplification functions, wavelet transform coefficient amplification, and response surface amplification. The exponential amplification function exponentially amplifies tiny deviations beyond the normal range, making the tiny changes more obvious; wavelet transform decomposes the signal at multiple scales, amplifies the wavelet coefficients representing abnormal features, and then reconstructs the signal; response surface amplification establishes a nonlinear relationship model between parameters and amplifies the coupling effect between parameters. After amplification processing, precursor signals such as weak temperature fluctuations, tension jitters, and gas concentration drifts become more obvious, facilitating subsequent identification and matching. Incorporating the typical evolution paths and characteristic sequences of different types of faults to form a characteristic time series pattern library for reflow soldering is an important link in establishing the knowledge basis for fault warning. This pattern library contains characteristic sequence templates for common fault types, and each template contains the parameter change sequences before, during, and after the occurrence of a fault. Typical faults include heating element aging faults, conveyor belt tension abnormal faults, gas system leakage faults, electrical control system faults, etc. For each fault, record the typical change patterns of parameters such as the temperature gradient distribution map, conveyor belt tension fluctuation characteristics, and gas concentration time-varying curve during its occurrence process, as well as the speed and severity of fault development. Each fault template in the pattern library contains information such as fault type, characteristic sequence, development speed, and severity, providing a reference standard for fault identification.

[0047] Calculating the similarity between the amplified fault symptom features and the time - series pattern library of reflow soldering features is the core step in identifying potential fault types. Multiple distance - metric methods are used for similarity calculation, including Euclidean distance, DTW distance, and Mahalanobis distance. Euclidean distance is suitable for comparing feature sequences of equal length; DTW distance can handle sequence comparisons with unequal lengths and time - axis distortions; Mahalanobis distance takes into account the correlation between features and is suitable for comprehensive comparisons of multi - dimensional features. For each possible fault type, calculate the similarity between the current amplified fault symptom features and the corresponding fault template in the pattern library to obtain a fault matching score. The matching score ranges from 0 to 1, and the larger the value, the higher the matching degree and the more likely it is to be this type of fault. Based on the fault matching score and the change trend of the equipment health index, predict the equipment state and possible fault risks at future time points. The prediction method combines pattern matching and time - series prediction techniques. First, determine the most likely fault type according to the fault matching score, then refer to the typical evolution path of this type of fault, and combine the historical change trend of the equipment health index to predict the future equipment state. The prediction results include possible fault types, fault occurrence probabilities, estimated occurrence times, and potential impact ranges. Finally, generate welding equipment fault warning information containing the above - mentioned information to provide guidance for equipment maintenance decisions.

[0048] Taking the reflow soldering equipment of an electronic manufacturing enterprise as an example, the equipment health monitoring system detected that the temperature fluctuation range in the middle of the reflow zone gradually increased from the normal ±2°C to ±3.5°C, and at the same time, the temperature gradient change rate increased slightly. These changes were recorded in the time - series mapping database. Through analysis, it was found that before the increase in temperature fluctuation, there were slight periodic fluctuations in the current of the heating element. Fault evolution pattern analysis identified this characteristic change rule: first, there were slight current fluctuations, then the temperature fluctuation increased, and finally the temperature gradient was abnormal and the health index decreased. After non - linear amplification processing of the current fluctuation signal, it was found that it presented pulsations with a frequency of 2 Hz and the amplitude gradually increased. Comparing this amplified feature with the pattern library, the matching degree with the "poor contact of heating element" fault mode was 0.87, indicating a relatively high probability of this fault. According to the typical evolution path of this fault, it was predicted that if no intervention was carried out, the equipment health index would drop from the current 0.85 to below 0.6 within the next 24 hours. A fault warning message was generated: risk of poor contact of heating element, matching degree 87%, expected to develop into moderate abnormality within 24 hours, it is recommended to check the contact of the heating element in the middle of the reflow zone after the end of today's production, enabling maintenance personnel to take timely intervention measures before the fault seriously affects production.

[0049] In a specific embodiment, the process of performing step S106 may specifically include the following steps: (1) Map and associate the welding equipment fault warning information with the equipment physical structure and operation logic to obtain equipment maintenance decision-making data; (2) Analyze the correlation between historical welding quality data and equipment status parameters to obtain a quality impact factor matrix; (3) Calculate the equipment downtime loss, maintenance labor cost, and spare part cost for different maintenance strategies to obtain a maintenance cost estimation table; (4) Use the quality impact factor matrix and the maintenance cost estimation table as constraint conditions, and calculate the optimal maintenance time window through an integer programming algorithm to obtain a maintenance time sequence arrangement; (5) Generate specific maintenance items and operation procedures based on the fault type in the welding equipment fault warning information and the equipment structure information to obtain a maintenance operation guide; (6) Integrate and process the maintenance time sequence arrangement and the maintenance operation guide to generate a welding equipment maintenance plan including maintenance time, maintenance items, required personnel, and resource allocation.

[0050] Specifically, map and associate the welding equipment fault warning information with the equipment physical structure and operation logic. This step transforms the abstract fault warning information into specific equipment components and functional units, and establishes a mapping relationship between the fault type and the maintenance object. First, according to the structure diagram and functional block diagram of the reflow welding equipment, the equipment is decomposed into main functional modules such as the heating system, conveyor belt system, gas system, and electrical control system. Then each module is further decomposed into specific components and parts. Through this hierarchical decomposition, a correspondence table between the fault type and the specific maintenance object is established. For example, a temperature anomaly fault may be mapped to a heating element, a temperature sensor, or a temperature controller; a conveyor belt anomaly fault may be mapped to a conveyor belt motor, a tension adjustment device, or a transmission gear. This mapping association transforms the fault warning information into clear maintenance decision-making data, including the specific components to be inspected, possible problem causes, and recommended handling methods. Analyzing the correlation between historical welding quality data and equipment status parameters is an important basis for optimizing maintenance strategies. By collecting welding quality inspection data (such as solder joint quality, welding strength, void ratio) over a period of time in the past and the corresponding equipment status parameters (such as temperature distribution, conveyor belt tension, gas concentration), statistical methods such as Pearson correlation coefficient and Spearman rank correlation coefficient are used to calculate the correlation between each parameter and the quality index. Based on the analysis results, a quality impact factor matrix is formed, and each element in the matrix represents the impact degree of a certain equipment parameter on a specific quality index. This matrix helps to determine which equipment components have the greatest impact on product quality, so as to prioritize the treatment of these key components when maintenance resources are limited.

[0051] Calculating the economic costs for different maintenance strategies is an important consideration in maintenance decision-making. Maintenance strategies include options such as immediate maintenance, deferred maintenance, scheduled downtime maintenance, and on-line partial maintenance. For each strategy, calculate three main types of costs: equipment downtime losses (including production delays, order postponements, and capacity losses), maintenance labor costs (including labor hours, overtime pay, and professional technical support fees), and spare parts costs (including component prices, transportation fees, and storage fees). Establish a cost model based on historical maintenance records and financial data, estimate the costs for different maintenance strategies, and form a maintenance cost estimate table to provide economic indicators for subsequent maintenance decisions. Use the quality impact factor matrix and the maintenance cost estimate table as constraints, and calculate the optimal maintenance time window through integer programming algorithms. The objective function of the integer programming model is to minimize the total cost (the sum of maintenance costs and quality loss costs), and the constraints include production plan requirements, resource availability, and maintenance time window limitations. By solving the integer programming problem, obtain the maintenance time arrangement with the lowest cost under the premise of meeting quality requirements, including the start time, end time, and timing relationship of maintenance.

[0052] Based on the fault type and equipment structure information in the welding equipment fault warning information, generate specific maintenance items and operation procedures. The maintenance items list in detail the specific components that need to be inspected, replaced, or adjusted, and the operation procedures describe the execution steps, required tools, and safety precautions of the maintenance activities. The maintenance operation guide usually includes steps for verifying the equipment functions to ensure that the equipment can operate normally and meet the design performance requirements after maintenance. Integrate the maintenance timing arrangement with the maintenance operation guide to generate a complete maintenance plan. The maintenance plan includes the maintenance time (date and period), maintenance items (specific components and treatment methods), required personnel (quantity, skill requirements), and resource allocation (tools, spare parts, equipment). The maintenance plan is presented in a structured document form for easy implementation management and progress tracking.

[0053] Taking an actual maintenance decision as an example, a fault warning system for a reflow soldering equipment detected a risk of poor contact of the heating element, with a matching degree of 87%. This warning information was mapped and associated with the equipment structure to determine that a specific heating element group in the middle of the reflow area and its electrical connection needed to be inspected. The quality impact analysis showed that the temperature stability in this area was highly correlated with the solder joint quality (correlation coefficient 0.82). The maintenance cost estimation showed that the cost of immediate shutdown for maintenance was high (including the delay of the current batch of products), while the cost could be reduced by 50% if the maintenance was planned after the end of today's production. Through integer programming calculation, the optimal maintenance time window was determined to be from 22:00 after the end of today's production to 2:00 am the next day. A maintenance operation guide was generated according to the fault type, including the disassembly steps of the heat dissipation system, the inspection method of the heating element, and the treatment process of the electrical contact points. The final generated maintenance plan included the specific time arrangement, the configuration of the maintenance team (1 electrical engineer and 2 maintenance technicians), and the list of required resources (contact cleaner, thermal conductive silicone grease, testing equipment, etc.), ensuring the efficient completion of the maintenance activities and eliminating the fault risk before the production impact occurred.

[0054] The operation status monitoring method of the reflow soldering equipment in the embodiment of the present application has been described above. Next, the operation status monitoring system of the reflow soldering equipment in the embodiment of the present application will be described. Please refer to Figure 2 , an embodiment of the operation status monitoring system of the reflow soldering equipment in the embodiment of the present application includes: An acquisition module, configured to acquire operation parameters through multi-point sensors of the reflow soldering equipment to obtain a welding equipment status data set; An association module, configured to extract a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve according to the welding equipment status data set, and perform multi-dimensional feature cross-correlation analysis to obtain a welding equipment operation feature matrix; A construction module, configured to construct a status recognition model based on the welding equipment operation feature matrix, classify the operation status of the equipment through a dynamic threshold adaptive adjustment mechanism, and obtain a welding equipment status category; A calculation module, configured to calculate an equipment health index according to the welding equipment status category, and set multi-level warning thresholds to obtain a welding equipment health assessment result; A comparison module, configured to construct a fault precursor feature extraction and amplification model by using the welding equipment health assessment result, and predict the future status through comparison with a reflow welding feature time series pattern library to obtain a welding equipment fault warning information; A generation module, configured to construct a multi-objective maintenance strategy optimization system based on the welding equipment fault warning information, and generate a maintenance plan through the correlation analysis of welding quality and equipment status to obtain a welding equipment maintenance plan.

[0055] Through the collaborative cooperation of the above-mentioned various components, a comprehensive dataset of the welding equipment status is obtained by collecting operation parameters through multi-point sensors, realizing the comprehensive monitoring of multiple systems and parameters of the equipment. Compared with the traditional single-point monitoring method, the comprehensiveness and accuracy of data collection are greatly improved, providing a rich data basis for subsequent analysis. Extract the temperature gradient distribution map, conveyor belt tension fluctuation characteristics, and gas concentration time-varying curve, and conduct multi-dimensional feature cross-correlation analysis to obtain the operation feature matrix of the welding equipment. This process fully explores the internal correlation between multi-dimensional parameters, reveals the mutual influence mechanism between different subsystems, and enables the abnormal state detection not to be limited to the judgment of a single parameter exceeding the limit, but to be evaluated from the overall performance of the system. The state recognition model constructed based on the operation feature matrix of the welding equipment realizes the accurate classification of the equipment operation state through a dynamic threshold adaptive adjustment mechanism. This mechanism dynamically adjusts the classification threshold according to the state transition frequency and duration in the equipment operation historical data, effectively solving the misjudgment problem caused by a fixed threshold, and improving the accuracy and stability of state recognition. Especially in the combined application of three machine learning algorithms, namely support vector machine, random forest, and gradient boosting decision tree, the respective advantages of different algorithms are fully utilized. The support vector machine is good at dealing with high-dimensional feature spaces and complex decision boundaries, the random forest has good anti-noise ability and feature importance evaluation ability, and the gradient boosting decision tree is good at dealing with imbalanced data and capturing weak feature changes. The weighted fusion of the three algorithms significantly improves the classification performance. Calculate the equipment health index according to the welding equipment state category and set multi-level warning thresholds, realizing the transformation from state classification to health quantification, making the equipment state evaluation more refined and intuitive. Use the welding equipment health assessment result to construct a fault precursor feature extraction and amplification model, and predict the future state by comparing with the reflow welding feature time series pattern library, transforming the passive response into an active prediction, greatly advancing the fault warning time, and gaining sufficient preparation time for maintenance decisions. Based on the welding equipment fault warning information, construct a multi-objective maintenance strategy optimization system, generate a maintenance plan through the correlation analysis of welding quality and equipment state, and introduce the product quality factor into the maintenance decision-making process for the first time, realizing the transformation from a simple equipment orientation to the coordinated optimization of quality and cost. This system solves for the optimal maintenance time window through an integer programming algorithm, minimizing the total maintenance cost on the premise of ensuring welding quality, providing scientific and economic maintenance decision support for enterprises, significantly improving production efficiency and product yield, and reducing maintenance costs and unplanned downtime losses.

[0056] Referring to Figure 3 , in the embodiment of the present invention, a computer device is further provided. This computer device can be a server, and its internal structure can be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected via a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0057] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0058] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0059] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0060] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, systems, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0061] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0062] As described above, the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or equivalently replace some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for monitoring the operating state of a reflow soldering device, characterized in that, The method for monitoring the operating state of the reflow soldering equipment includes: Collecting operating parameters through multi-point sensors of the reflow soldering equipment to obtain a soldering equipment state data set; Extracting a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the soldering equipment state data set, and performing multi-dimensional feature cross-correlation analysis to obtain a soldering equipment operation feature matrix; Constructing a state recognition model based on the soldering equipment operation feature matrix, classifying the equipment operating state through a dynamic threshold adaptive adjustment mechanism to obtain the soldering equipment state category; Calculating the equipment health index according to the soldering equipment state category, and setting multi-level warning thresholds to obtain the soldering equipment health assessment result; Using the soldering equipment health assessment result to construct a fault precursor feature extraction and amplification model, and predicting the future state by comparing with the reflow soldering feature time series pattern library to obtain the soldering equipment fault warning information; Constructing a multi-objective maintenance strategy optimization system based on the soldering equipment fault warning information, and generating a maintenance plan through the correlation analysis of welding quality and equipment state to obtain the soldering equipment maintenance plan.

2. The method for monitoring the operating state of a reflow soldering device according to claim 1, characterized in that, The collecting operating parameters through multi-point sensors of the reflow soldering equipment to obtain a soldering equipment state data set includes: Installing temperature sensor arrays in the preheating zone, reflow zone, and cooling zone of the reflow soldering equipment, installing speed sensors and tension sensors in the conveyor belt system, installing gas concentration sensors and air pressure sensors in the gas supply system, and installing current and voltage sensors in the power supply system to form a multi-type sensor network; Using an industrial Internet of Things module to uniformly sample the data collected by the multi-type sensor network, where the temperature data sampling frequency is 10Hz, the conveyor belt parameter sampling frequency is 5Hz, the gas parameter sampling frequency is 2Hz, and the electrical parameter sampling frequency is 20Hz; Denosing the collected temperature data using a moving average filtering algorithm, denosing the conveyor belt speed data using a median filtering algorithm, and processing the gas concentration data using a Kalman filtering algorithm; Completing the missing data points using linear interpolation or polynomial interpolation according to their continuous missing lengths, and preliminarily screening the outliers using the statistical 3σ principle; Aligning the processed data of various sensors according to a unified time stamp to generate a soldering equipment state data set.

3. The method for monitoring the operating state of a reflow soldering device according to claim 1, wherein The extracting a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the soldering equipment state data set, and performing multi-dimensional feature cross-correlation analysis to obtain a soldering equipment operation feature matrix includes: Segmenting the soldering equipment state data set according to different time window lengths, and the window length is adaptively adjusted to 5 minutes to 30 minutes according to different equipment operation cycles; Calculating the temperature gradient distribution map for the temperature data within each time window, obtaining the spatial temperature gradient by dividing the difference between adjacent temperature sensors by the sensor spacing, and then combining the time dimension to form a temperature gradient spatio-temporal distribution matrix; Extract the conveyor belt tension fluctuation characteristics from the conveyor belt parameters, including the peak-to-peak value of tension, the tension stability coefficient, and the tension pulsation frequency characteristics, and construct a tension fluctuation feature vector; Perform time-frequency transformation on the gas concentration data, and extract the time-varying curve characteristics of the gas concentration, including the gas concentration change rate, the gas concentration fluctuation period, and the proportion of the gas concentration stable interval; Calculate the phase correlation degree between the temperature gradient distribution map and the conveyor belt tension fluctuation characteristics, analyze the time-delay correlation between the gas concentration time-varying curve and the temperature gradient, and construct a parameter cross-influence matrix; Perform dimensionality reduction processing on the extracted features through the principal component analysis method, retain the principal components with a cumulative contribution rate reaching 95%, and form an operating characteristic matrix of the welding equipment.

4. The method for monitoring the operating state of a reflow soldering device according to claim 1, wherein, Construct a state recognition model based on the operating characteristic matrix of the welding equipment, classify the operating state of the equipment through a dynamic threshold adaptive adjustment mechanism, and obtain the welding equipment state categories, including: Pre-define five typical operating states of the reflow welding equipment: normal operating state, temperature abnormal state, conveyor belt abnormal state, gas system abnormal state, and electrical system abnormal state; Divide the operating characteristic matrix of the welding equipment into a training matrix and a verification matrix, and the training matrix is used for subsequent classification algorithm training; Process the training matrix through the support vector machine algorithm, construct a first classifier, and evaluate the performance of the first classifier with the verification matrix to obtain the first classification accuracy; Process the training matrix through the random forest algorithm, construct a second classifier, and evaluate the performance of the second classifier with the verification matrix to obtain the second classification accuracy; Process the training matrix through the gradient boosting decision tree algorithm, construct a third classifier, and evaluate the performance of the third classifier with the verification matrix to obtain the third classification accuracy; Construct a dynamic threshold adaptive adjustment mechanism, and dynamically adjust the discriminant threshold parameters of the first classifier, the second classifier, and the third classifier according to the state transition frequency and duration in the historical operating data of the equipment; According to the first classification accuracy, the second classification accuracy, and the third classification accuracy, fuse the discriminant results of the three classifiers through the weighted voting method to obtain the welding equipment state categories.

5. The method for monitoring the operating state of a reflow soldering device according to claim 1, wherein, Calculate the equipment health index according to the welding equipment state categories, and set multi-level warning thresholds to obtain the welding equipment health assessment results, including: Convert the welding equipment state categories into subsystem state quantization values; For the temperature abnormal state, conveyor belt abnormal state, gas system abnormal state, and electrical system abnormal state in the welding equipment state categories, assign subsystem weight coefficients to obtain the weighted parameters of the subsystem state quantization values; Multiply the subsystem state quantization value by the subsystem weight coefficient to obtain a weighted state score; Extract the abnormal duration characteristics from the historical changes of the welding equipment state categories to obtain an abnormal duration factor; Quantify the degree of abnormal diffusion in the system based on the relevance of multiple subsystem abnormalities in the welding equipment state categories to obtain an abnormal diffusion factor; Comprehensively calculate the weighted status score, the abnormal duration factor, and the abnormal diffusion factor to obtain the equipment health index, and set multi-level warning thresholds according to the equipment health index to generate the health assessment result of the welding equipment.

6. The method for monitoring the operating state of a reflow soldering device according to claim 1, characterized in that, Use the health assessment result of the welding equipment to construct a fault precursor feature extraction and amplification model, and predict the future state by comparing with the reflow welding feature time series pattern library to obtain the fault warning information of the welding equipment, including: Associate the health assessment result of the welding equipment with the equipment parameters at historical time points to obtain a time series mapping database; Extract the fault evolution pattern from the time series mapping database, identify the characteristic change law before the decline of the equipment health index, and obtain the fault precursor feature set; Perform non-linear amplification processing on the weak change signals in the fault precursor feature set to obtain the amplified fault symptom features; Collect the typical evolution paths and characteristic sequences of different types of faults to form a reflow welding feature time series pattern library; Calculate the similarity between the amplified fault symptom features and the reflow welding feature time series pattern library, identify the potential fault types and development trends, and obtain the fault matching degree score; Based on the fault matching degree score and the change trend of the equipment health index, predict the equipment state at future time points and the possible fault risks, and generate the fault warning information of the welding equipment.

7. The method for monitoring the operating state of a reflow soldering device according to claim 1, characterized in that, Based on the fault warning information of the welding equipment, construct a multi-objective maintenance strategy optimization system, and generate a maintenance plan through the correlation analysis of welding quality and equipment state to obtain the maintenance plan of the welding equipment, including: Map and associate the fault warning information of the welding equipment with the equipment physical structure and operation logic to obtain equipment maintenance decision data; Analyze the correlation between historical welding quality data and equipment state parameters to obtain a quality impact factor matrix; Calculate the equipment downtime loss, maintenance labor cost, and spare parts cost for different maintenance strategies to obtain a maintenance cost estimation table; Use the quality impact factor matrix and the maintenance cost estimation table as constraint conditions, and calculate the optimal maintenance time window through an integer programming algorithm to obtain the maintenance time sequence arrangement; Generate specific maintenance items and operation procedures according to the fault types and equipment structure information in the fault warning information of the welding equipment to obtain a maintenance operation guide; Integrate the maintenance time sequence arrangement and the maintenance operation guide to generate a maintenance plan for the welding equipment including maintenance time, maintenance items, required personnel, and resource allocation.

8. A monitoring system for the operating state of a reflow soldering device, which is used to implement the method for monitoring the operating state of a reflow soldering device according to any one of claims 1-7, characterized in that, The reflow welding equipment operation status monitoring system includes: An acquisition module for acquiring operation parameters through multi-point sensors of the reflow welding equipment to obtain a welding equipment state data set; An association module for extracting a temperature gradient distribution map, a conveyor belt tension fluctuation feature, and a gas concentration time-varying curve from the welding equipment state data set, and performing multi-dimensional feature cross-correlation analysis to obtain a welding equipment operation feature matrix; A construction module for constructing a state recognition model based on the welding equipment operation feature matrix, and classifying the equipment operation status through a dynamic threshold adaptive adjustment mechanism to obtain the welding equipment state category; A calculation module, configured to calculate an equipment health index according to the welding equipment status category, set multi-level warning thresholds, and obtain a welding equipment health assessment result; A comparison module, configured to use the welding equipment health assessment result to construct a fault precursor feature extraction and amplification model, compare and predict the future state through a reflow welding feature time series pattern library, and obtain welding equipment fault warning information; A generation module, configured to construct a multi-objective maintenance strategy optimization system based on the welding equipment fault warning information, generate a maintenance plan through the correlation analysis of welding quality and equipment status, and obtain a welding equipment maintenance plan.

9. A computer device, characterized in that, It includes a memory and a processor. The memory stores a computer program that can run on the processor. It is characterized in that when the processor executes the computer program, the method for monitoring the operating state of the reflow welding equipment according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the processor is caused to execute the method for monitoring the operating state of the reflow welding equipment according to any one of claims 1 to 7.

Images