Industrial equipment multi-parameter collaborative measurement and control method and system

Abstract

The application belongs to the field of automation measurement and control technology, and is a kind of multi-parameter collaborative measurement and control method and system for industrial equipment, which comprises the following steps: obtaining multi-source operation information and performing dimensionless processing; calculating the short-time signal-to-noise ratio of the multi-source operation information under different sensor channels; comparing each short-time signal-to-noise ratio with the ideal signal-to-noise ratio benchmark; when the short-time signal-to-noise ratio of any channel is less than the ideal signal-to-noise ratio benchmark corresponding to the channel, adjusting the weight distribution of all channel sensors by using an exponential decay function; performing weighted fusion operation on the current signal values of all sensor channels according to the adjusted fusion weight, and outputting a fusion signal; extracting the change trend feature of the fusion signal and inputting the change trend feature into an abnormality discrimination model; when the abnormality discrimination model outputs an abnormal state identifier, generating and sending an execution action instruction for the industrial equipment; the application has high measurement and control precision and strong anti-interference ability, and breaks through the limitations of traditional fixed weight through dynamic self-adaptive weighted fusion.

Description

Technical Field

[0001] This application belongs to the field of automated measurement and control technology, specifically a multi-parameter collaborative measurement and control method and system for industrial equipment. Background Technology

[0002] With the development of intelligent manufacturing in industry, the monitoring of the operating status of key industrial equipment such as machine tools, pumps and valves, and automated production lines has become particularly important. In order to comprehensively obtain the equipment status, existing measurement and control systems are usually equipped with multiple sensors (such as vibration, temperature, pressure, and speed sensors), and multi-sensor data fusion technology is used to improve the accuracy and reliability of measurement and control.

[0003] Currently, existing technologies for multi-parameter measurement and control of industrial equipment mainly fall into the following two categories:

[0004] In terms of data processing: static weight allocation or conventional Kalman filtering algorithm is commonly used. That is, during system initialization, fixed weight values ​​are preset according to the calibration accuracy of the sensors, or state estimation is performed based on the assumption of ideal Gaussian white noise.

[0005] In terms of system architecture: a traditional architecture is typically adopted, consisting of sensor acquisition - PLC / host computer display - manual / threshold control. Sensors are responsible for collecting data, which is transmitted to the central controller via an industrial bus for simple threshold comparison.

[0006] Although the aforementioned existing technologies have achieved equipment monitoring to a certain extent, they still have the following significant shortcomings when facing complex industrial field conditions, which urgently need to be addressed:

[0007] First, traditional fusion algorithms cannot handle non-Gaussian sudden interference. The standard Kalman filter algorithm strictly relies on the ideal assumption that measurement noise follows Gaussian white noise. However, in actual industrial settings, due to factors such as the start-up and shutdown of high-power motors and arc discharge, environmental noise often exhibits non-Gaussian impulse noise or a heavy-tailed distribution. When such sudden strong electromagnetic interference causes a sensor's signal-to-noise ratio (SNR) to drop sharply, the standard Kalman filter cannot identify this outlier and still includes it in the update process, leading to estimation divergence or severe bias due to model mismatch. Simultaneously, the static weighting method cannot automatically eliminate bad points when the sensor experiences zero-point drift or failure, severely impacting measurement and control accuracy.

[0008] Second, monitoring and control are disconnected, lacking the ability to predict faults. Traditional architectures are mostly reactive, i.e., "fault-alarm-shutdown," lacking a fault trend prediction mechanism based on multi-dimensional data characteristics. In addition, the drift calibration of sensor nodes relies on regular manual maintenance and cannot form a millisecond-level "sensing-control-closed loop" with the actuators, resulting in a persistently high equipment failure downtime rate.

[0009] Third, interface compatibility is poor, and the transformation cost is high. The hardware interfaces of traditional measurement and control systems are mostly proprietary protocols or fixed interfaces of various brands, which cannot be directly compatible with sensors and actuators from different manufacturers (such as simultaneously supporting mainstream protocols such as RS485, Modbus, and Profinet). When enterprises carry out intelligent transformation, they often need to add a large number of adapter devices, resulting in low installation and debugging efficiency, lack of hot-swapping support, and difficult maintenance. Summary of the Invention

[0010] This application provides a method and system for multi-parameter collaborative measurement and control of industrial equipment, which solves the problems of weak anti-interference ability, disconnect between monitoring and control, and poor interface compatibility in existing industrial measurement and control systems.

[0011] The first aspect of this application provides a method for multi-parameter collaborative measurement and control of industrial equipment, the method comprising:

[0012] Multi-source operational information is acquired and dimensionless processing is performed. The multi-source operational information comes from sensor signals collected by different sensor nodes. The sampling frequency is adjusted by judging the operating conditions of industrial equipment.

[0013] Calculate the short-time signal-to-noise ratio of the multi-source operational information under different sensor channels;

[0014] Each short-time signal-to-noise ratio is compared with the ideal signal-to-noise ratio benchmark. When the short-time signal-to-noise ratio of any channel is less than the ideal signal-to-noise ratio benchmark corresponding to that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function.

[0015] Based on the adjusted fusion weights, a weighted fusion operation is performed on the current signal values ​​of all sensor channels, and the fused signal is output.

[0016] Extract the trend features of the fused signal and input the trend features into the anomaly detection model;

[0017] When the anomaly detection model outputs an anomaly status identifier, it generates and sends an execution action command for the industrial equipment.

[0018] Furthermore, the self-learning and initialization of benchmark feature parameters before acquiring multi-source operational information includes:

[0019] Obtain sensor configuration parameters for industrial equipment;

[0020] Based on the sensor configuration parameters, a logical acquisition channel with a unique identifier is established for each sensor;

[0021] The real-time operating status of industrial equipment is acquired, and when the operating status indicates that the equipment is in rated operating condition, a data acquisition trigger signal is generated.

[0022] In response to the acquisition trigger signal, the system receives the raw signals transmitted from the multi-channel sensors and splits the raw signals according to the unique identifier of the logical acquisition channel.

[0023] The split signal streams are reassembled in chronological order and stored as discrete time-series datasets named after their channel numbers.

[0024] Furthermore, the self-learning and initialization of the benchmark feature parameters also includes:

[0025] Based on the time-series sampled values ​​in the discretized time-series dataset fed back by each sensor within a predetermined period, a mean statistical feature quantity characterizing the central tendency of the periodic signal and a square root statistical feature quantity characterizing the fluctuation degree of the periodic signal are generated, and the mean statistical feature quantity and the square root statistical feature quantity are set as the reference benchmark for subsequent signal-to-noise ratio calculation.

[0026] Further, the short-time signal-to-noise ratio of the multi-source operational information under different sensor channels is calculated, including:

[0027] Obtain a reference benchmark for each sensor channel, and perform dimensionless standardization processing on the real-time signal values ​​of each sensor channel based on the reference benchmark to output a standardized signal sequence.

[0028] A sliding time window buffer with a fixed length is instantiated for each sensor channel, and the standardized signal sequence is continuously pushed into the corresponding sliding time window in chronological order to form a window dataset that is updated in real time.

[0029] Based on the dataset within the sliding time window at the current moment, calculate the signal mean and signal variance of the signal within the sliding time window respectively;

[0030] The short-time signal-to-noise ratio of the channel at the current moment is generated by a logarithmic domain transformation based on the ratio of the square of the signal mean to the signal variance.

[0031] Furthermore, each short-time signal-to-noise ratio (SNR) is compared with an ideal SNR benchmark. When the short-time SNR of any channel is less than the corresponding ideal SNR benchmark, the weight allocation of all channel sensors is adjusted using an exponential decay function, including:

[0032] Calculate the exponential decay factor for the sensor corresponding to each channel;

[0033] The normalized weights are calculated based on the exponential decay factor of the sensor corresponding to each channel, ensuring that the sum of the weights of each channel is 1.

[0034] Further, the exponential decay factor of the sensor corresponding to each channel is calculated, including:

[0035] Calculate the absolute value of the absolute deviation between the signal-to-noise ratio of the i-th channel and the ideal signal-to-noise ratio reference, and normalize the absolute value of the absolute deviation to the [0,1] interval to obtain the relative deviation index of the signal-to-noise ratio.

[0036] Introducing a sensitivity adjustment coefficient , This is a preset positive real number used to control the penalty strength of the relative signal-to-noise ratio deviation index on weight attenuation;

[0037] A nonlinear mapping model based on an exponential decay function is constructed to map the relative deviation index of the signal-to-noise ratio to an exponential decay factor in the interval [0,1]. The nonlinear mapping model is expressed as follows:

[0038] ,

[0039] in For the first The exponential decay factor of the channel, As an ideal signal-to-noise ratio benchmark, For the first aisle Signal-to-noise ratio at any given moment.

[0040] Furthermore, the anomaly detection model predicts potential failure risks in the following manner:

[0041] Calculate the minimum cumulative distance between the current trend feature vector and the historical normal operating condition trend template library, and trigger an early warning when the cumulative distance exceeds the early warning threshold;

[0042] Extract the reconstruction probability of the trend feature sequence, and determine the trend as abnormal when the reconstruction probability is lower than the preset quantile;

[0043] The probability distribution of trend feature values ​​at the next moment is predicted in real time, and a prediction is triggered when the measured value deviates from the 95% confidence interval.

[0044] Furthermore, after predicting potential fault risks, the anomaly detection model uses a parallel OR logic method to monitor the following two types of triggering conditions in real time:

[0045] Triggering condition A: Calculate the instantaneous amplitude characteristics or envelope amplitude characteristics of the fused signal in real time. When the amplitude characteristics exceed the preset safety threshold, generate an over-limit trigger flag.

[0046] Triggering condition B: Real-time tracking of the temporal evolution trajectory of the exponential decay factor of each channel to construct a sensor failure judgment criterion: When the exponential decay factor of any i-th channel is continuously lower than the failure judgment threshold for a period of time exceeding the failure confirmation window length, it is determined that the sensor of that channel has suffered a substantial failure or signal link interruption, and a failure triggering identifier is generated.

[0047] Furthermore, the generation and transmission of execution action instructions for industrial equipment includes:

[0048] Receive the anomaly status identifier output by the anomaly discrimination model, wherein the anomaly status identifier includes: anomaly type code, anomaly confidence level, and anomaly evolution trend;

[0049] Based on the current abnormal status identifier and the real-time collected industrial equipment operating parameters, matching association rules are retrieved from the knowledge base to generate instantiated execution action instructions. The knowledge base is a pre-built association mapping knowledge base for industrial equipment fault modes and execution action instructions.

[0050] A second aspect of this application provides an industrial equipment multi-parameter collaborative measurement and control system, comprising:

[0051] A multi-channel sensor array is used to acquire multi-source operating information and perform dimensionless processing. The multi-source operating information comes from sensor signals collected by different sensor nodes. The sampling frequency is adjusted by judging the operating conditions of industrial equipment.

[0052] The short-time signal-to-noise ratio calculation module is used to calculate the short-time signal-to-noise ratio of the multi-source operating information under different sensor channels;

[0053] The weighting module compares each short-time signal-to-noise ratio (SNR) with the ideal SNR benchmark. When the short-time SNR of any channel is less than the ideal SNR benchmark for that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function.

[0054] The fusion module is used to perform a weighted fusion operation on the current signal values ​​of all sensor channels based on the adjusted fusion weights, and output the fused signal.

[0055] The feature extraction module is used to extract the trend features of the fused signal and input the trend features into the anomaly detection model;

[0056] The anomaly detection module is used to generate and send execution action instructions for industrial equipment when an anomaly status identifier is output.

[0057] Compared with existing technologies, the advantages of this application are as follows: the method of this application has high measurement and control accuracy and strong anti-interference ability. Through dynamic adaptive weighted fusion, it breaks through the limitations of traditional fixed weights. In a strong interference environment, the multi-parameter fusion measurement and control error is controlled within ±0.8%, which is more than 22% higher than the accuracy of the traditional static fusion scheme and 30% higher than the electromagnetic interference resistance.

[0058] This solved the problem of the disconnect between monitoring and control. By predicting potential faults, proactive prevention and control were achieved, reducing equipment downtime by 18%. Attached Figure Description

[0059] Figure 1 A flowchart of a multi-parameter collaborative measurement and control method for industrial equipment provided in this application embodiment;

[0060] Figure 2 A flowchart illustrating the method for self-learning and initializing baseline feature parameters provided in this application embodiment;

[0061] Figure 3 A block diagram of a multi-parameter collaborative measurement and control system for industrial equipment provided in the embodiments of this application;

[0062] Figure 4 The following are comparison diagrams of the results under dynamic and complex working conditions provided in the embodiments of this application with those of existing technical methods: (a) comparison diagram when handling sudden interference, (b) comparison diagram when capturing signal-to-noise ratio degradation, and (c) final measurement and control comparison diagram. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] In actual industrial settings, environmental noise from high-power motors starting and stopping, arcing, and other factors often manifests as non-Gaussian impulse noise or heavy-tailed noise. When such sudden, strong electromagnetic interference causes a sensor's signal-to-noise ratio to drop sharply, standard Kalman filtering fails to identify outliers and still includes them in the update process, leading to estimation divergence or severe bias due to model mismatch. Furthermore, static weighting methods cannot automatically eliminate dead pixels when sensors experience zero-point drift or malfunction, seriously affecting measurement and control accuracy.

[0065] This application provides a multi-parameter collaborative measurement and control method and system for industrial equipment. It calculates the short-time signal-to-noise ratio (SNR) of different channels by collecting multi-source operational information from different nodes using different sensors. The SNR is then compared to a preset ideal SNR benchmark. When the SNR of any channel is less than the corresponding ideal SNR benchmark, an exponential decay function is used to adjust the weight allocation of all channel sensors. A weighted fusion operation is then performed on the current signal value based on the updated weights. The resulting fused signal is then analyzed for anomalies based on its trend characteristics. The industrial equipment then performs corresponding actions according to the type of anomaly. This dynamic adaptive weighted fusion overcomes the limitations of traditional fixed weights, thereby improving the electromagnetic interference resistance of industrial equipment and reducing downtime.

[0066] The industrial equipment applicable to the embodiments of this application can be: rotating equipment, such as motors, fans, water pumps, compressors, turbines, generators, gearboxes, and bearings; reciprocating mechanical equipment, such as piston compressors, internal combustion engines, plunger pumps, stamping presses, and forging presses; fluid conveying equipment, such as various pumps (centrifugal pumps, gear pumps, screw pumps), valves, pipeline systems, and hydraulic systems; process control equipment, such as regulating valves, actuators, transmitters, positioners, and DCS / PLC systems; and robots and automation equipment, such as industrial robots, AGVs, automated production lines, CNC machine tools, and 3D printers.

[0067] See Figure 1 The method for multi-parameter collaborative measurement and control of industrial equipment shown includes:

[0068] S101: Acquire multi-source operating information and perform dimensionless processing. The multi-source operating information comes from sensor signals collected by different sensor nodes. Multi-source operating parameters (including vibration, temperature, pressure, and speed) of the industrial equipment are collected through a multi-protocol compatible sensing interface. The collected data undergoes automatic zero-point drift compensation through a built-in reference signal module or reference signal generator in the sensor. The calculation frequency and sampling frequency are automatically adjusted based on the operating condition of the industrial equipment (no-load / load / standby). When the industrial equipment is determined to be in no-load or standby state, the sampling frequency is automatically reduced (e.g., from 100Hz to 20Hz) to reduce power consumption. For example, when the current of the industrial equipment is detected to be less than 10% of the rated current (no-load / standby), the sampling frequency is automatically reduced from 100Hz to 20Hz to reduce energy consumption.

[0069] S102, Calculate the short-time signal-to-noise ratio of the multi-source operation information under different sensor channels;

[0070] S103 compares each short-time signal-to-noise ratio with the ideal signal-to-noise ratio benchmark. When the short-time signal-to-noise ratio of any channel is less than the ideal signal-to-noise ratio benchmark corresponding to that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function.

[0071] S104, Based on the adjusted fusion weights, perform a weighted fusion operation on the current signal values ​​of all sensor channels and output the fused signal;

[0072] S105, extract the trend features of the fused signal and input the trend features into the anomaly detection model;

[0073] S106, when the anomaly discrimination model outputs an anomaly status identifier, an execution action instruction for the industrial equipment is generated and sent.

[0074] The specific implementation methods of each step in the embodiments of this application are described in detail below.

[0075] Multi-source operational information from industrial equipment is acquired through a multi-protocol compatible sensing interface, and the acquired data undergoes automatic zero-point drift compensation via a built-in reference signal module. The multi-protocol compatible sensing interface enables interoperability between sensors from different manufacturers and of different types. The reference signal module is a precise and stable reference source embedded within the sensor. Its core function is to provide a known zero-point or full-scale reference signal that does not change with external measured variables and environmental conditions (such as temperature and time). Automatic compensation using the reference signal module is a typical system self-calibration process. The deviation between the reference and zero point is measured periodically, and the measured values ​​are corrected in real time accordingly.

[0076] Multi-source operational information can include, but is not limited to, vibration, temperature, pressure, and rotational speed. This information can be acquired using the same or different types of sensors, such as vibration sensors, temperature sensors, pressure sensors, and rotational speed sensors.

[0077] Multi-source operational information is acquired using the same or different sensors. Multiple sensors are installed at different nodes of industrial equipment, such as machine tool spindles, pumps, valves, and pipelines. The sensor nodes adopt a modular design and acquire vibration, temperature, pressure, and speed signals through standard interfaces (supporting hot-swapping of RS485 / Modbus / Profinet protocols). Due to the differences in industrial equipment, the number and type of sensors used vary. To adapt to the individual differences of different industrial equipment (such as different models of machine tools or pumps and valves), baseline parameters need to be initialized and entered before formal operation.

[0078] In one embodiment, see Figure 2 As shown, self-learning and initialization of baseline feature parameters are performed before acquiring multi-source operational information. This is to adapt to the individual differences of different industrial equipment (such as different models of machine tools or pumps and valves), including:

[0079] S201, Obtain sensor configuration parameters for industrial equipment. Sensor configuration parameters include sensor type, installation location, measurement range, and sampling frequency information.

[0080] S202, based on the sensor configuration parameters, a logical acquisition channel with a unique identifier is established for each sensor, forming a one-to-one mapping relationship between the logical acquisition channel and the physical sensor. In the core processing unit of the data acquisition and monitoring system, based on the sensor configuration parameters parsed in step S201, a corresponding logical acquisition channel object is created for each physical sensor. This process specifically includes: dynamically allocating channel data structures in system memory and assigning a globally unique channel identifier to each channel object. This channel identifier is not only bound to the sensor's MAC address or hardware interface number, but also forms a mapping relationship with the spatial topology location of the device, thereby establishing a precise correspondence between the physical world and the digital world.

[0081] S203: Acquire the real-time operating status of the industrial equipment. When the operating status indicates that the equipment is in its rated operating condition, generate a data acquisition trigger signal. In one application, the operating parameters of the industrial equipment are read in real time through a programmable logic controller (PLC) or a distributed control system. For example, the monitored operating parameters include the motor load rate, spindle speed, critical bearing temperature, and housing vibration intensity. The system has a preset rated operating condition threshold range table. For example, the rated operating condition is defined as: load rate between 85% and 115%, speed within ±5% of the rated speed, and temperature below a set alarm value. When all the real-time read operating parameters fall within this threshold range, the equipment is determined to be in a stable rated operating condition.

[0082] S204, in response to the acquisition trigger signal, the original signal returned by the multi-channel sensor is received, and the original signal is split according to the unique identifier of the logical acquisition channel; in response to the acquisition trigger signal generated in step S203, the multi-channel data acquisition card starts working in parallel scanning mode.

[0083] S205 reassembles the split signal streams in time order and stores them as discrete time-series datasets named after the channel numbers.

[0084] By triggering data acquisition under rated operating conditions, characteristic data of industrial equipment in stable operating conditions can be obtained, avoiding the impact of operating condition fluctuations on data quality. By establishing logical acquisition channels with unique identifiers and performing split processing on the raw signals, effective isolation and organization of multi-source heterogeneous data are achieved. By recombining the signal stream into a discrete time-series dataset named with channel numbers, a structured data foundation is provided for subsequent equipment status analysis, fault diagnosis, and health assessment, improving data utilization efficiency and analysis accuracy.

[0085] In addition, the self-learning and initialization conditions for triggering the reference characteristic parameters can also be: when the system is first installed, the sensor is replaced, or a manual "reset calibration" command is received, it enters "self-learning mode". At this time, the operator must ensure that the industrial equipment is in a "healthy and stable rated operating state". A high frequency (such as 100Hz) can be selected to continuously collect sensor data for a period of time (e.g., 30~60 seconds).

[0086] The baseline characteristic parameters are calculated and stored, including:

[0087] Based on the time-series sampled values ​​from the discretized time-series dataset fed back by each sensor within a predetermined period, a mean statistical feature representing the central tendency of the periodic signal and a square root variance statistical feature representing the degree of fluctuation of the periodic signal are generated. These mean and square root variance statistical features are then set as reference benchmarks for subsequent signal-to-noise ratio calculations. These mean and square root variance statistical features are permanently stored in the non-volatile memory of the edge computing gateway.

[0088] The mean statistical characteristic is the arithmetic mean of the data from each sensor; the square root of the variance statistical characteristic refers to the standard deviation. The mean statistical characteristic and the square root of the variance statistical characteristic are used as the benchmark mean and benchmark standard deviation, respectively.

[0089] Because the collected signals, such as vibration, temperature, and pressure, have inconsistent physical dimensions, they cannot be directly fused mathematically. Therefore, the system calls a reference benchmark and uses the Z-score method to perform dimensionless processing on the real-time data.

[0090] For the One sensor in Data collected at any given time Its standardized value The calculation is as follows:

[0091] ,

[0092] in: This represents the baseline average value of the sensor when the device is in a healthy state.

[0093] This represents the baseline standard deviation of the sensor under device health conditions.

[0094] Physical meaning: Standardized value This represents the degree to which the current signal deviates from a normal healthy state. If... A value close to 0 indicates that the equipment is operating smoothly; if the absolute value is much greater than 3, it indicates that the current state has deviated significantly from the baseline operating condition.

[0095] Based on the reference benchmark, the real-time signal values ​​of each sensor channel are processed into dimensionless normalization, and a normalized signal sequence is output to calculate the short-time signal-to-noise ratio of multi-source operating information under different sensor channels.

[0096] Based on the multi-source operation information in step S101, the short-time signal-to-noise ratio of the multi-source operation information under different sensor channels is calculated in real time. The calculation of the short-time signal-to-noise ratio is based on the signal-to-noise ratio of the mean and variance of the signal within the sliding window.

[0097] For each sensor signal channel, a sliding window of length N is established. The sliding window is updated in real time with the sampling points to always keep it up-to-date. Each sampled data point, within the current sliding window. Calculate the arithmetic mean of the sampled data. As an estimate of the effective signal components within that time period, the variance of the data within the current sliding window is calculated. As an energy estimate of the noise components within this time period, based on the mean... and variance Calculate the signal-to-noise ratio of the current sliding window.

[0098] A sliding window technique is used to perform real-time analysis of the preprocessed standardized signal sequence. Specifically, the system maintains a sliding window of length for each sensor channel. First-in-first-out (FIFO) data cache queue.

[0099] The choice of the sliding window length N is crucial to the algorithm's performance. To achieve a balance between anti-interference sensitivity and estimation stability, the window length in this embodiment... The preferred setting is 50-100 sampling points (taking a 100Hz sampling rate as an example, corresponding to a time span of 0.5 seconds to 1.0 second). If the sliding window is too short (e.g.) The calculated signal-to-noise ratio (SNR) value can fluctuate drastically due to minute fluctuations in the signal itself; if the sliding window is too long (e.g., ...), the SNR value will also fluctuate drastically due to minute fluctuations in the signal itself. If so, the response to sudden interference will be significantly delayed.

[0100] At each time t, based on the dataset within the current sliding window, calculate the short-time signal-to-noise ratio of the i-th sensor. The calculation formula is as follows:

[0101] ,

[0102] in: The mean of the signal within the sliding window represents the effective trend component of the signal.

[0103] The variance of the signal within the sliding window represents the discrete noise intensity of the signal.

[0104] For regularization parameters (in this embodiment, we take...) This is used to prevent program overflow when the denominator is zero when the sensor signal is completely constant (variance is 0).

[0105] Based on the aforementioned short-time signal-to-noise ratios (SNRs), each SNR is compared to an ideal SNR benchmark. When the SNR of any channel is less than the corresponding ideal SNR benchmark, the weight allocation of all channel sensors is adjusted using an exponential decay function, including:

[0106] Calculate the exponential decay factor for the sensor corresponding to each channel;

[0107] The normalized weights are calculated based on the exponential decay factor of the sensor corresponding to each channel, ensuring that the sum of the weights of each channel is 1.

[0108] Calculate the exponential decay factor for each channel's corresponding sensor, including:

[0109] Calculate the absolute value of the absolute deviation between the signal-to-noise ratio of the i-th channel and the ideal signal-to-noise ratio reference, and normalize the absolute value of the absolute deviation to the [0,1] interval to obtain the relative deviation index of the signal-to-noise ratio.

[0110] Introducing a sensitivity adjustment coefficient , This is a preset positive real number used to control the penalty strength of the relative signal-to-noise ratio deviation index on weight attenuation;

[0111] A nonlinear mapping model based on an exponential decay function is constructed to map the relative deviation index of the signal-to-noise ratio to an exponential decay factor in the interval [0,1]. The nonlinear mapping model is expressed as follows:

[0112] ,

[0113] in For the first The exponential decay factor of the channel, As an ideal signal-to-noise ratio benchmark, For the first aisle Signal-to-noise ratio at any given moment.

[0114] Based on a nonlinear mapping model with an exponential decay function, dynamic weight iteration is achieved. The system sets an ideal signal-to-noise ratio benchmark. When a decrease in the signal-to-noise ratio of a certain sensor is detected, the weight ratio of that sensor is automatically reduced using the exponential decay function, while the weight of high-reliability sensors is increased. The final weighted fusion measurement and control value is then calculated.

[0115] Ideal Signal-to-Noise Ratio (SNR) Reference Setting: For each type of sensor or each specific channel, set an ideal SNR reference value. This ideal SNR reference value can be determined based on: the sensor's nominal performance parameters, the long-term average SNR measured under good conditions during the system calibration phase, and the median value of the optimal operating range obtained from statistical analysis of historical data, etc.

[0116] Assign initial fusion weights to each sensor. This can be done using equal weights or non-equal weights based on prior knowledge. Key parameter settings for the dynamic weight iterative model:

[0117] The higher the value of the forgetting factor, the smoother the weight changes and the stronger the system's ability to resist short-term interference; the lower the value, the faster the weights respond to changes in the signal-to-noise ratio.

[0118] Attenuation coefficient: Used to control the rate at which the weight decays with the signal-to-noise ratio gap. It is usually determined through calibration.

[0119] Weight lower limit: To prevent the sensor from completely failing due to temporary interference, a non-zero minimum weight is set. When the calculated weight is lower than this value, it is forcibly assigned the weight lower limit to preserve its recovery possibility.

[0120] By weighting sensor signals using an exponential decay factor, adaptive weight allocation based on signal-to-noise ratio can be achieved during multi-sensor fusion, effectively suppressing interference contributions from low signal-to-noise ratio channels and improving the robustness and output accuracy of the fusion system in complex environments.

[0121] Final normalized weights The calculation is as follows:

[0122] ,

[0123] The final output value of the fused signal .

[0124] Extract the trend characteristics of the fused signal, realize fault prediction and coordinated control through the anomaly discrimination model, predict the fault risk based on the equipment's operating characteristics, and synchronously trigger the execution command to fine-tune parameters (such as cooling, speed reduction) or perform emergency shutdown.

[0125] Specifically, the anomaly detection model uses the following methods to predict potential failure risks:

[0126] Calculate the minimum cumulative distance between the current trend feature vector and the historical normal operating condition trend template library. When the cumulative distance exceeds the early warning threshold, trigger an early warning. The trend feature vector refers to an array of multi-dimensional mathematical features extracted from the fused data sink that can represent the parameters "direction of change", "rate of change" and "shape of change" to quantify the dynamic change law of the operating status of industrial equipment.

[0127] The reconstruction probability of the trend feature sequence is extracted. When the reconstruction probability is lower than a preset quantile, the trend is considered abnormal. The trend feature sequence refers to the features formed by arranging the operating trends of industrial equipment within a continuous time window in chronological order. The reconstruction probability refers to inputting the current trend feature sequence into a pre-trained deep neural network model (such as an autoencoder, variational autoencoder, or generative adversarial network). The deep neural network model attempts to encode and then decode (i.e., reconstruct) the input sequence based on its normal pattern memory learned during training. The reconstruction probability measures the degree of similarity between the decoded sequence and the original input sequence, or the probability density in the model's latent space that the current sample belongs to a normal distribution. The preset quantile is a threshold boundary set after statistically analyzing the reconstruction probability values ​​under all historical normal operating conditions. The preset quantile serves as the statistical boundary for judging normal and abnormal conditions. Trend abnormality means that the current equipment operating state deviates from the statistical regularity of historical normal operating conditions, indicating that the equipment may have early failures or abnormal operating conditions.

[0128] The probability distribution of trend feature values ​​at the next moment is predicted in real time, and a prediction is triggered when the measured value deviates from the 95% confidence interval.

[0129] After anticipating potential failure risks, the following two types of triggering conditions are monitored in real time using a parallel OR logic connection:

[0130] Triggering condition A: Calculate the instantaneous amplitude characteristics or envelope amplitude characteristics of the fused signal in real time. When the amplitude characteristics exceed the preset safety threshold, generate an over-limit trigger flag.

[0131] Triggering condition B: Real-time tracking of the temporal evolution trajectory of the exponential decay factor of each channel to construct a sensor failure judgment criterion: When the exponential decay factor of any i-th channel is continuously lower than the failure judgment threshold for a period of time exceeding the failure confirmation window length, it is determined that the sensor of that channel has suffered a substantial failure or signal link interruption, and a failure triggering identifier is generated.

[0132] When the triggering conditions are met, an action command for the industrial equipment is generated and sent.

[0133] In one embodiment, based on the current abnormal state identifier and the real-time collected industrial equipment operating parameters, matching association rules are retrieved from a knowledge base to generate instantiated execution action instructions. The knowledge base is a pre-constructed association mapping knowledge base for industrial equipment fault modes and execution action instructions. The abnormal state identifier includes: anomaly type code, anomaly confidence level, and anomaly evolution trend. Anomaly type code: a code used to uniquely identify the fault or anomaly category, such as "BEARING-OVHT-001" representing bearing overheating. Anomaly confidence level: characterizes the reliability of the current anomaly determination result, with a value range of [0,1], where a higher value indicates a more reliable detection result. Anomaly evolution trend: describes the dynamic changes of the abnormal state over time, and may include discrete states such as "rapid deterioration," "slow development," or "tending to stabilize," or continuous values ​​quantified by trend feature vectors. An association mapping knowledge base is constructed, pre-constructing an association mapping knowledge base for industrial equipment fault modes and execution action instructions. The knowledge base is constructed by collecting historical fault cases and expert handling records, and then using natural language processing and association rule mining algorithms to extract the mapping relationship between fault patterns (including anomaly types and operating conditions) and handling actions. Each rule in the knowledge base consists of three core parts:

[0134] Rule preconditions (conditions): specific anomaly type and corresponding operating condition parameter threshold range;

[0135] Rule consequent (action): A standardized template for executing action instructions;

[0136] Rule attributes: including the rule's confidence level, support level, and applicable device models.

[0137] Retrieve and match association rules: Using the obtained current abnormal status identifier and real-time operating parameters as retrieval conditions, semantic matching and parameter matching are performed in the association mapping knowledge base constructed in step S2 to retrieve the set of association rules that meet the triggering conditions. The specific matching process includes: first, performing precise matching or fuzzy matching based on the abnormal type code; then, comparing each candidate rule in the matching set to see if the current operating parameters meet the operating conditions set in the rule antecedent (e.g., "current load > 80%)". Finally, generate instantiated execution action instructions. Based on the retrieved association rules that meet the matching conditions, the execution action instruction template in the rules is combined with the current abnormal status identifier and real-time operating parameters to perform parameter instantiation and filling, generating specific instantiated execution action instructions that can be directly parsed by the downstream execution system. The instantiated execution action instructions include at least: target actuator address, action type (e.g., "adjustment", "stop", "alarm"), action parameters (e.g., adjustment range, alarm level), and execution time limit.

[0138] Secondly, see Figure 3As shown, this application also provides an industrial equipment multi-parameter collaborative measurement and control system applying the above method, including: a multi-channel sensor array for acquiring multi-source operating information and performing dimensionless processing, wherein the multi-source operating information comes from sensor signals collected by different sensor nodes, and the sampling frequency is adjusted by judging the operating condition of the industrial equipment;

[0139] The short-time signal-to-noise ratio calculation module is used to calculate the short-time signal-to-noise ratio of the multi-source operating information under different sensor channels;

[0140] The weighting module compares each short-time signal-to-noise ratio (SNR) with the ideal SNR benchmark. When the short-time SNR of any channel is less than the ideal SNR benchmark for that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function.

[0141] The fusion module is used to perform a weighted fusion operation on the current signal values ​​of all sensor channels based on the adjusted fusion weights, and output the fused signal.

[0142] The feature extraction module is used to extract the trend features of the fused signal and input the trend features into the anomaly detection model;

[0143] The anomaly detection module is used to generate and send execution action instructions for industrial equipment when an anomaly status identifier is output.

[0144] The architecture includes:

[0145] The sensing layer comprises several sensor nodes with self-calibration capabilities, supporting multiple communication protocols including RS485, Modbus, and Profinet, and featuring hot-swappable interfaces. It acquires multi-source operational information from sensor signals collected by different sensor nodes and is deployed in critical components such as machine tool spindles, pumps, valves, and pipelines. The sensor nodes employ a modular design, acquiring vibration, temperature, pressure, and speed signals via standard interfaces (supporting hot-swappable RS485 / Modbus / Profinet protocols). Each sensor integrates a reference signal generator, periodically producing a standard level signal for zero-point drift self-calibration.

[0146] Edge processing layer: Employs a high-performance embedded processor and deploys a short-time signal-to-noise ratio calculation module, a weighting module, a fusion module, a feature extraction module, and an anomaly detection module for real-time computation and decision-making of local data.

[0147] The execution control layer includes frequency converters, cooling pump controllers, etc. It receives control commands from the edge layer via the industrial bus and execution commands from the edge processing layer to precisely control the actuators of industrial equipment.

[0148] To fully verify the effectiveness of this invention under dynamic and complex working conditions and demonstrate its significant advantages over existing technologies, this embodiment constructs a simulation model containing three redundant sensors. The actual physical quantity is set as a dynamic sinusoidal signal with a frequency of 0.5 Hz to simulate the rapid changes in the state of industrial equipment during actual operation. To simulate typical harsh working conditions in industrial settings, non-Gaussian burst interference (broadband noise intensity amplified 16 times) is injected into sensor 1 during the simulation time t=10s to 12s to simulate the effects of arc discharge or the start-up and shutdown of high-power equipment. The experiment also introduces the "traditional arithmetic mean method" and the "traditional Kalman filter" as control groups. The traditional Kalman filter, to ensure the tracking capability of the 0.5 Hz dynamic signal, specifically sets a high process noise covariance Q.

[0149] See Figure 4 As shown, the experimental results clearly demonstrate the superiority of this application in handling sudden interference. See also Figure 4 In (a), there are three sensors: sensor 1, sensor 2, and sensor 3. Sensor 1 is set to be subject to interference, while sensors 2 and 4 are normal. The raw signals collected by the three sensors show that sensor 1 suffers from strong non-Gaussian interference. At the moment the interference occurs, the method of this application quickly detects the deterioration in the signal-to-noise ratio, as shown in the results. Figure 4 In (b), the weights of the three sensors are w1 (sensor 1), w2 (sensor 2), and w3 (sensor 3). An exponential decay model is used to vertically drop the weight w1 of sensor 1 to 0 within milliseconds, achieving soft cutoff of fault data. This is reflected in the final measurement and control effect, see [see...]. Figure 4 In (c), the true value, the averaging method, the traditional Kalman method, and the present application are compared. The output curve of the traditional averaging method (arithmetic average method) oscillates violently during the interference period and is completely distorted. Although the traditional Kalman method has a certain filtering effect, it falls into the inherent contradiction of "sensitivity and anti-interference" because its measurement noise matrix R is fixed. The sensitivity that is increased in order to track dynamic signals makes it unable to effectively isolate sudden noise, and the output results still show significant deviation and fluctuation.

[0150] In contrast, the fused output curve of this application closely follows the true value benchmark throughout the entire interference period, without significant oscillations. This simulation strongly demonstrates that by introducing the signal-to-noise ratio as an adaptive variable, this application successfully solves the technical challenge of balancing "dynamic signal tracking" and "non-Gaussian burst interference suppression" in traditional methods. While retaining its ability to keenly capture rapidly changing equipment states, the system achieves perfect immunity to burst interference under harsh operating conditions, exhibiting extremely high robustness.

[0151] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for multi-parameter collaborative measurement and control of industrial equipment, characterized in that, The method includes: Multi-source operational information is acquired and dimensionless processing is performed. The multi-source operational information comes from sensor signals collected by different sensor nodes. The sampling frequency is adjusted by judging the operating conditions of industrial equipment. Calculate the short-time signal-to-noise ratio of the multi-source operational information under different sensor channels; Each short-time signal-to-noise ratio is compared with the ideal signal-to-noise ratio benchmark. When the short-time signal-to-noise ratio of any channel is less than the ideal signal-to-noise ratio benchmark corresponding to that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function. include: Calculate the exponential decay factor for the sensor corresponding to each channel; Calculate the normalized weight based on the exponential decay factor of the sensor corresponding to each channel, ensuring that the sum of the weights of each channel is 1; Calculate the exponential decay factor for each channel's corresponding sensor, including: Calculate the absolute value of the absolute deviation between the signal-to-noise ratio of the i-th channel and the ideal signal-to-noise ratio reference, and normalize the absolute value of the absolute deviation to the [0,1] interval to obtain the relative deviation index of the signal-to-noise ratio. Introducing a sensitivity adjustment coefficient , This is a preset positive real number used to control the penalty strength of the relative signal-to-noise ratio deviation index on weight attenuation; A nonlinear mapping model based on an exponential decay function is constructed to map the relative deviation index of the signal-to-noise ratio to an exponential decay factor in the interval [0,1]. The nonlinear mapping model is expressed as follows: ， in For the first The exponential decay factor of the channel, As an ideal signal-to-noise ratio benchmark, For the first aisle Signal-to-noise ratio at any given moment; Based on the adjusted fusion weights, a weighted fusion operation is performed on the current signal values ​​of all sensor channels, and the fused signal is output. Extract the trend features of the fused signal and input the trend features into the anomaly detection model; When the anomaly detection model outputs an anomaly status identifier, it generates and sends an execution action command for the industrial equipment.

2. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 1, characterized in that, The self-learning and initialization of baseline feature parameters before acquiring multi-source operational information includes: Obtain sensor configuration parameters for industrial equipment; Based on the sensor configuration parameters, a logical acquisition channel with a unique identifier is established for each sensor; The real-time operating status of industrial equipment is acquired, and when the operating status indicates that the equipment is in rated operating condition, a data acquisition trigger signal is generated. In response to the acquisition trigger signal, the system receives the raw signals transmitted from the multi-channel sensors and splits the raw signals according to the unique identifier of the logical acquisition channel. The split signal streams are reassembled in chronological order and stored as discrete time-series datasets named after their channel numbers.

3. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 2, characterized in that, The self-learning and initialization of the benchmark feature parameters also includes: Based on the time-series sampled values ​​in the discretized time-series dataset fed back by each sensor within a predetermined period, a mean statistical feature quantity characterizing the central tendency of the periodic signal and a square root statistical feature quantity characterizing the fluctuation degree of the periodic signal are generated, and the mean statistical feature quantity and the square root statistical feature quantity are set as the reference benchmark for subsequent signal-to-noise ratio calculation.

4. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 3, characterized in that, Calculating the short-time signal-to-noise ratio of the multi-source operational information under different sensor channels includes: Obtain a reference benchmark for each sensor channel, and perform dimensionless standardization processing on the real-time signal values ​​of each sensor channel based on the reference benchmark to output a standardized signal sequence. A sliding time window buffer with a fixed length is instantiated for each sensor channel, and the standardized signal sequence is continuously pushed into the corresponding sliding time window in chronological order to form a window dataset that is updated in real time. Based on the dataset within the sliding time window at the current moment, calculate the signal mean and signal variance of the signal within the sliding time window respectively; The short-time signal-to-noise ratio of the channel at the current moment is generated by a logarithmic domain transformation based on the ratio of the square of the signal mean to the signal variance.

5. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 1, characterized in that, The anomaly detection model predicts potential failure risks using the following methods: Calculate the minimum cumulative distance between the current trend feature vector and the historical normal operating condition trend template library, and trigger an early warning when the cumulative distance exceeds the early warning threshold; Extract the reconstruction probability of the trend feature sequence, and determine the trend as abnormal when the reconstruction probability is lower than the preset quantile; The probability distribution of trend feature values ​​at the next moment is predicted in real time, and a prediction is triggered when the measured value deviates from the 95% confidence interval.

6. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 5, characterized in that, After predicting potential fault risks, the anomaly detection model uses a parallel OR logic method to monitor the following two types of triggering conditions in real time: Triggering condition A: Calculate the instantaneous amplitude characteristics or envelope amplitude characteristics of the fused signal in real time. When the amplitude characteristics exceed the preset safety threshold, generate an over-limit trigger flag. Triggering condition B: Real-time tracking of the temporal evolution trajectory of the exponential decay factor of each channel to construct a sensor failure judgment criterion: When the exponential decay factor of any i-th channel is continuously lower than the failure judgment threshold for a period of time exceeding the failure confirmation window length, it is determined that the sensor of that channel has suffered a substantial failure or signal link interruption, and a failure triggering identifier is generated.

7. The method for multi-parameter collaborative measurement and control of industrial equipment according to claim 6, characterized in that, The process of generating and sending execution instructions for industrial equipment includes: Receive the anomaly status identifier output by the anomaly discrimination model, wherein the anomaly status identifier includes: anomaly type code, anomaly confidence level, and anomaly evolution trend; Based on the current abnormal status identifier and the real-time collected industrial equipment operating parameters, matching association rules are retrieved from the knowledge base to generate instantiated execution action instructions. The knowledge base is a pre-built association mapping knowledge base for industrial equipment fault modes and execution action instructions.

8. A multi-parameter collaborative measurement and control system for industrial equipment, used to execute the method according to any one of claims 1-7, characterized in that, include: A multi-channel sensor array is used to acquire multi-source operating information and perform dimensionless processing. The multi-source operating information comes from sensor signals collected by different sensor nodes. The sampling frequency is adjusted by judging the operating conditions of industrial equipment. The short-time signal-to-noise ratio calculation module is used to calculate the short-time signal-to-noise ratio of the multi-source operating information under different sensor channels; The weighting module compares each short-time signal-to-noise ratio (SNR) with the ideal SNR benchmark. When the short-time SNR of any channel is less than the ideal SNR benchmark for that channel, the weight allocation of all channel sensors is adjusted using an exponential decay function. The fusion module is used to perform a weighted fusion operation on the current signal values ​​of all sensor channels based on the adjusted fusion weights, and output the fused signal. The feature extraction module is used to extract the trend features of the fused signal and input the trend features into the anomaly detection model; The anomaly detection module is used to generate and send execution action instructions for industrial equipment when an anomaly status identifier is output.

Images