An artificial intelligence operation and maintenance management method and system based on an internet of things

Abstract

The application relates to the technical field, and particularly discloses an artificial intelligence operation and maintenance management method and system based on an Internet of Things, which collects a device position identification code to call corresponding ground rigidity coefficients and settlement accumulations, establishes a dynamic baseline model containing ground parameters and operation condition parameters for each device, differentiates vibration differential values by differentiating real-time vibration data from predicted baseline values output by the model to eliminate ground fixed deviation, collects environmental temperature, humidity and adjacent device states, separates environmental interference components from the vibration differential values through regression analysis to obtain fault characteristic components, and finally outputs a health state evaluation result, thereby solving the problem of misjudging the health degree of the device caused by the fixed deviation of the vibration baseline due to ground differences.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance technology, and specifically to an artificial intelligence operation and maintenance management method and system based on the Internet of Things. Background Technology

[0002] In the equipment operation and maintenance management of the intelligent manufacturing field, existing technologies typically collect vibration data by placing vibration sensors on the equipment, extract feature values ​​by combining them with operating parameters such as temperature and current, and input them into an artificial intelligence model to assess the health status of the equipment.

[0003] However, the foundations of equipment installed in different locations within the same production workshop exhibit physical characteristics such as uneven settlement and differences in rigidity. When the same type of equipment operates under different foundation conditions, its foundation vibration values ​​and noise baselines naturally show fixed deviations. Existing operation and maintenance management methods, when extracting vibration characteristics for health scoring, fail to effectively separate this vibration deviation introduced by the physical differences in the foundation from the actual fault characteristics of the equipment. As a result, equipment with slightly softer foundations and slightly higher vibration baselines is judged by artificial intelligence models to have extremely poor health and is thus prioritized for maintenance, while equipment with genuine potential faults is overlooked, leading to misallocation of operation and maintenance resources and missed fault detection. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based operation and maintenance management method and system based on the Internet of Things (IoT) to solve the aforementioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: An AI-based operation and maintenance management method based on the Internet of Things includes the following steps: Collect basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; A dynamic baseline model for each piece of equipment is established using the equipment installation environment feature set as input. The dynamic baseline model outputs the predicted baseline value of the foundation vibration data. The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; Collect environmental interference parameters at the location of each device, and merge the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. The equipment health status assessment results are output based on the magnitude and trend of the fault characteristic components.

[0006] As a further aspect of the present invention: collecting basic vibration data, equipment operating condition parameters, and location identification codes of each piece of equipment in the production workshop includes: During equipment installation, triaxial accelerometers are placed at the four corners of the contact surface between the equipment base and the foundation. The triaxial accelerometers are synchronized with the clock of the central control system in the production workshop. The triaxial accelerometers continuously collect the vibration acceleration time-domain signal of the equipment under no-load operation. The vibration acceleration time-domain signal is Fourier transformed to obtain the frequency domain amplitude spectrum. The amplitude corresponding to the natural frequency of the equipment structure is extracted from the frequency domain amplitude spectrum and recorded as the basic vibration data. The equipment controller reads the spindle speed, feed speed, and load current of the equipment in real time and records them as operating parameters. The coordinate code stored in the RFID tag embedded in the foundation is used as the location identification code of the equipment, and the reading range of the RFID tag is limited to the coverage area of ​​the equipment installation base.

[0007] As a further aspect of the present invention: the extraction of foundation response characteristic parameters includes: The foundation monitoring database pre-stores the foundation rigidity coefficient and cumulative foundation settlement corresponding to each location identifier. The IoT gateway retrieves the foundation rigidity coefficient and cumulative foundation settlement from the foundation monitoring database using the location identifier as an index and records them as foundation response characteristic parameters. The foundation rigidity coefficient is obtained by conducting an impact excitation test on the foundation before equipment installation. During the impact excitation test, a standard pulse load is applied to the center point of the equipment installation location, and a displacement sensor is placed at a preset distance from the center point. The peak displacement response is recorded by the displacement sensor, and the ratio of the amplitude of the standard pulse load to the peak displacement response is defined as the foundation rigidity coefficient. The cumulative foundation settlement is obtained by continuously collecting data using a static level instrument placed around the equipment installation location. The elevation change value is recorded every 24 hours by the static level instrument, and the difference between the elevation change value and the initial installation elevation is accumulated to obtain the cumulative foundation settlement.

[0008] As a further aspect of the present invention: obtaining the predicted baseline value includes: The foundation rigidity coefficient and cumulative foundation settlement, which are part of the equipment installation environment feature set, are used as fixed inputs to the model's structural parameters. The equipment operating condition parameters are used as dynamic inputs to the model, and the equipment foundation vibration data are used as the model's output target. A nonlinear regression algorithm is used to fit the response surface of the foundation vibration data as the equipment operating condition parameters change within the operating condition parameter space. This response surface is recorded as the equipment's dynamic baseline model, with each operating condition parameter point on the response surface corresponding to a predicted baseline value. The equipment's unique identification code is associated with and stored in relation to the function expression of the response surface. Each time a new cumulative foundation settlement is obtained through a hydrostatic level, the nonlinear regression algorithm is re-executed to fit the response surface.

[0009] As a further aspect of the present invention: calculating the vibration difference value includes: The equipment controller reads the spindle speed, feed rate, and load current in real time and inputs these values ​​as query coordinates into the equipment dynamic baseline model. The dynamic baseline model then performs trilinear interpolation on the response surface based on the query coordinates to obtain the predicted baseline value. A triaxial accelerometer collects the vibration acceleration time-domain signal in real time and converts it into an effective vibration value. The effective vibration value is then subtracted from the predicted baseline value to obtain the vibration difference value. The sign of the vibration difference value indicates the direction of deviation of the effective vibration value from the predicted baseline value, and the absolute value of the vibration difference value indicates the magnitude of the deviation.

[0010] As a further aspect of the present invention: separating environmental interference components includes: The regression analyzer pre-stores a set of historical fusion feature vectors for all equipment in the workshop. This set includes vibration difference values ​​and corresponding ambient temperature, humidity, and operating status parameters of nearby equipment. The ambient temperature, humidity, and operating status parameters are denoted as environmental disturbance parameters. The regression analyzer establishes a multiple linear regression equation with the vibration difference value as the dependent variable and the ambient temperature, humidity, and operating status parameters as independent variables. It then uses the least squares method to solve for the partial regression coefficients of each environmental variable in the multiple linear regression equation. The environmental disturbance component is obtained by summing the products of the ambient temperature value and its corresponding partial regression coefficient, the ambient humidity value and its corresponding partial regression coefficient, and the operating status parameters of nearby equipment and their corresponding partial regression coefficients. The fault characteristic component is obtained by subtracting the environmental disturbance component from the vibration difference value.

[0011] As a further aspect of the present invention: the output device health status assessment results include: The fault feature components are arranged in order of their acquisition timestamps to form a fault feature component time series. A moving window smoothing process is then applied to this time series to obtain a smoothed fault feature component time series. The current value of the smoothed fault feature component time series is compared with a preset warning threshold. When the current value exceeds the warning threshold, the slope change rate of the smoothed fault feature component time series over the past continuous acquisition periods is extracted. If the slope change rate shows an increasing trend, a first-level warning indicator is output; if the slope change rate shows a decreasing trend, a second-level warning indicator is output. The equipment health status assessment results, the equipment unique identifier, the location identifier, and the fault feature component time series are packaged and stored in the operation and maintenance management database.

[0012] An IoT-based AI-powered operation and maintenance management system includes: Initial module: Collects basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; Prediction module: The equipment dynamic baseline model is established for each piece of equipment using the equipment installation environment feature set as input. The equipment dynamic baseline model outputs the predicted baseline value of the foundation vibration data. The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; Optimization module: Collects environmental interference parameters at the location of each device, and merges the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. Management module: Outputs equipment health status assessment results based on the magnitude and trend of fault characteristic components.

[0013] The beneficial effects of this invention compared to the prior art are as follows: This invention collects the location identification code of the equipment and retrieves the foundation rigidity coefficient and cumulative foundation settlement at the corresponding location. An independent dynamic baseline model is established for each piece of equipment. This model outputs predicted baseline values ​​for foundation vibration data under arbitrary operating conditions. The real-time collected foundation vibration data is then compared with the predicted baseline values ​​to obtain a vibration difference value, eliminating the fixed bias caused by differences in foundation rigidity and settlement. Furthermore, the invention collects ambient temperature, humidity, and operating status parameters of nearby equipment at the equipment location. A regression analyzer separates the environmental interference component introduced by dynamic environmental disturbance parameters from the vibration difference value, defining the remaining portion as fault characteristic components. This invention separates foundation physical differences, dynamic environmental disturbances, and actual equipment fault characteristics layer by layer, ensuring that the final output fault characteristic components only reflect the equipment's own fault state. This avoids misjudgments caused by foundation differences and environmental interference, improving the accuracy of equipment health status assessment. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a flowchart illustrating an AI-based operation and maintenance management method based on the Internet of Things according to the present invention. Figure 2 This is a schematic diagram of the process for calculating the predicted baseline value according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figures 1-2 As shown, this invention is an AI-based operation and maintenance management method based on the Internet of Things, comprising the following steps: Collect basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; In a preferred embodiment of the present invention, during equipment installation, triaxial accelerometers are respectively arranged at the four corners of the contact surface between the equipment base and the foundation. The four corners are referenced to the geometric center of the rectangular plane of the equipment base, extending to the four corners of the base. Each triaxial accelerometer is connected to a distributed data acquisition terminal via a data acquisition cable, and the distributed data acquisition terminal is connected to the central control system via Ethernet. The central control system operates a Network Time Protocol (NTP) service, and the distributed data acquisition terminal initiates a time synchronization request to the central control system and adjusts its local clock to ensure that the data acquisition time of each triaxial accelerometer is consistent with the clock of the central control system.

[0018] In the no-load state, the central control system sends a continuous acquisition command to the distributed data acquisition terminal. The distributed data acquisition terminal simultaneously activates the triaxial accelerometers at four corner points at a sampling frequency of 25.6 kHz, continuously acquiring vibration acceleration time-domain signals for 10 seconds. The triaxial accelerometers convert the vibration acceleration time-domain signals into analog voltage signals, which are then stored in a buffer in time-series format after analog-to-digital conversion. The central control system performs a Fast Fourier Transform on the vibration acceleration time-domain signals at each corner point to obtain the frequency domain amplitude spectrum. It locates the frequency point corresponding to the natural frequency of the equipment structure in the frequency domain amplitude spectrum and extracts the amplitude at that frequency point. For example, if the natural frequency of the spindle box assembly of a certain model of machining center is 128 Hz, the central control system extracts the amplitude at 128 Hz as the basic vibration data of that corner point. The arithmetic mean of the basic vibration data of the four corner points is taken as the basic vibration data of the equipment.

[0019] Equipment operating parameters are acquired through the equipment controller. The equipment controller is a programmable logic controller (PLC) connected to the spindle driver, feed servo driver, and current transformer via a fieldbus. The PLC sends read commands to the spindle driver to obtain spindle speed values, such as 8000 rpm; it sends read commands to the feed servo driver to obtain feed speed values, such as 500 mm / min; and it obtains load current values, such as 45 amps, through an analog input module connected to the secondary side of the current transformer. The PLC packages and transmits these three parameters—spindle speed, feed speed, and load current—to the central control system, which records them as operating parameters.

[0020] The fieldbus connects to the spindle driver, feed servo driver, and current transformer. The programmable logic controller (PLC) sends read commands to the spindle driver to obtain the spindle speed, for example, 8000 rpm; it sends read commands to the feed servo driver to obtain the feed rate, for example, 500 mm / min; and it obtains the load current, for example, 45 amps, by connecting an analog input module to the secondary side of the current transformer. The PLC packages and transmits the three parameters—spindle speed, feed rate, and load current—to the central control system, which records them as operating condition parameters.

[0021] The location identification code of the equipment is obtained through RFID tags pre-embedded in the foundation. During foundation pouring, the RFID tags are buried 20 cm below the foundation surface. Each RFID tag stores a coordinate code, for example, location A03. After the equipment is installed, an RFID reader is installed on the side of the equipment base. The reader's reading range is limited to the area covered by the equipment mounting base, with a reading radius set to 30 cm. The central control system periodically sends reading commands to the RFID reader. The reader activates the RFID tag and reads the stored coordinate code, returning it to the central control system, which records it as the location identification code. The central control system then binds the foundation vibration data, operating parameters, and location identification code according to the acquisition timestamp and uploads them to the IoT gateway.

[0022] The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; In a preferred embodiment of the present invention, the extraction of foundation response characteristic parameters includes: After receiving the location identifier, the IoT gateway performs a retrieval operation in the foundation monitoring database using the location identifier as an index. The foundation monitoring database adopts a relational database management system, and the database tables have fields for location identifier, foundation rigidity coefficient, and cumulative foundation settlement. The IoT gateway constructs a structured query language statement, passes the location identifier as a query condition into the statement, and after executing the query, extracts the foundation rigidity coefficient and cumulative foundation settlement from the returned result set. The two extracted parameters are recorded as foundation response characteristic parameters.

[0023] The foundation stiffness coefficient was obtained through impact excitation testing before equipment installation. An impact hammer was placed at the center point of the equipment installation location. The hammerhead of the impact hammer had a built-in force sensor, which was connected to a dynamic signal analyzer via a signal line. The impact hammer struck the foundation surface vertically downwards, generating a pulse load with a duration of 0.2 milliseconds and a peak amplitude of 10 kN. The force sensor transmitted the time-domain waveform of the pulse load to the dynamic signal analyzer. An eddy current displacement sensor was placed 1.5 meters away from the center point. The probe of the eddy current displacement sensor was vertically aligned with the foundation surface, and the initial gap between the probe and the foundation surface was set to 2 mm. When the impact load was applied, the eddy current displacement sensor recorded the time-domain waveform of the displacement response of the foundation surface. The dynamic signal analyzer simultaneously received the pulse load waveform from the force sensor and the displacement response waveform from the displacement sensor, extracting the peak displacement response from the displacement response waveform. For example, in one test, the peak displacement response was 0.05 mm. The dynamic signal analyzer divides the pulse load amplitude of 10 kN by the peak displacement response of 0.05 mm, and defines the calculated result of 200 kN / mm as the foundation stiffness coefficient at that location.

[0024] The cumulative ground settlement was continuously collected using hydrostatic levels. Four hydrostatic levels were arranged around the equipment installation location, connected in series via a connecting pipe to form a communicating vessel system. Each hydrostatic level contained an ultrasonic level sensor. Upon completion of installation, the initial liquid level height of each hydrostatic level was recorded, and this elevation value was recorded as the initial installation elevation. The hydrostatic levels automatically performed a measurement every 24 hours. The ultrasonic level sensor emitted ultrasonic pulses to the liquid surface, and the time difference between the emission and the reflected echo was measured. The current liquid level height was calculated based on this time difference, and the elevation change was obtained by subtracting the current liquid level height from the initial installation elevation. For example, if a hydrostatic level measured on the 30th day after installation, the current liquid level height was 125 mm, the initial installation elevation was 120 mm, and the elevation change was 5 mm. The daily elevation changes are added to the historical cumulative values. For example, if the elevation change is 0.2 mm on day 1, 0.3 mm on day 2, and 0.2 mm on day 3, the cumulative change over day 30 is 5 mm. This cumulative value is recorded as the cumulative foundation settlement. The cumulative foundation settlement measured by each hydrostatic level is uploaded to the foundation monitoring database via a wireless transmission module and stored in association with the corresponding location identifier code.

[0025] A dynamic baseline model for each piece of equipment is established using the equipment installation environment feature set as input. The dynamic baseline model outputs the predicted baseline value of the foundation vibration data. In a preferred embodiment of the present invention, obtaining the predicted baseline value includes: The IoT gateway transmits the equipment installation environment feature set to the central processing unit (CPU). The CPU extracts the foundation rigidity coefficient and cumulative foundation settlement from this feature set, using these parameters as structural parameters for the equipment's dynamic baseline model. Simultaneously, the CPU extracts equipment operating condition parameters from the feature set, including spindle speed, feed rate, and load current, which are used as dynamic input variables for the dynamic baseline model. Finally, the CPU extracts foundation vibration data from the feature set, using this data as the output target variable for the dynamic baseline model.

[0026] The central processing unit (CPU) calls a nonlinear regression algorithm package, which employs a Gaussian process regression algorithm. The CPU inputs the foundation stiffness coefficient and cumulative foundation settlement as fixed parameters into the Gaussian process regression algorithm, and uses the spindle speed, feed rate, and load current as three-dimensional coordinate points in the input space. Foundation vibration data is used as the target values ​​corresponding to each coordinate point in the input space. The CPU collects data samples of the equipment under multiple operating conditions. For example, the foundation vibration data corresponding to a spindle speed of 8000 rpm, a feed rate of 500 mm / min, and a load current of 45 amps is 0.15 mm / s; the corresponding data for a spindle speed of 12000 rpm, a feed rate of 800 mm / min, and a load current of 60 amps is 0.22 mm / s; and the corresponding data for a spindle speed of 4000 rpm, a feed rate of 200 mm / min, and a load current of 25 amps is 0.09 mm / s. The central processing unit (CPU) inputs these data samples as a training set into the Gaussian process regression algorithm. The Gaussian process regression algorithm fits a continuous hypersurface in the three-dimensional working condition parameter space consisting of spindle speed, feed rate, and load current. Each three-dimensional coordinate point on the hypersurface corresponds to a predicted basic vibration value. The CPU records this hypersurface as the equipment dynamic baseline model and stores the function expression of the hypersurface in the model library in the form of a mathematical formula.

[0027] In this method, the foundation stiffness coefficient and the cumulative foundation settlement are used as fixed parameters input into the Gaussian process regression algorithm. The Gaussian process regression algorithm uses fixed parameters by concatenating these fixed parameters with the operating condition parameters as additional input dimensions during the training phase to form the input vector. When processing the input vector, the kernel function of the Gaussian process regression algorithm treats the dimensions corresponding to these fixed parameters as constant dimensions. During the calculation of the covariance matrix, the algorithm sets the difference in these constant dimensions to zero, so that the fixed parameters are only used as conditional variables to adjust the overall offset of the output response surface. This method of inputting environmental physical parameters as covariates into the Gaussian process is a common technique in the field of engineering modeling and will not be elaborated here. The central processing unit (CPU) assigns a unique device identification code to each device, using a 16-bit hexadecimal encoding, for example, 2A3F8C9E1B4D6F7H. The CPU associates the device identification code with the function expression of the device's dynamic baseline model in a mapping table within the model library. The mapping table uses the device identification code as the primary key and the text form of the function expression as the value. Each time the hydrostatic level acquires a new cumulative foundation settlement, the foundation monitoring database pushes an update notification to the IoT gateway, which then forwards the update notification to the CPU. The central processing unit (CPU) retrieves the foundation stiffness coefficient and updated cumulative foundation settlement from the foundation monitoring database based on the equipment identification code. Simultaneously, it extracts all operating condition parameters and corresponding foundation vibration data accumulated by the equipment during the update cycle from the historical database. The CPU uses the re-acquired foundation stiffness coefficient, cumulative foundation settlement, and newly accumulated operating condition parameters and foundation vibration data as a new training set, and re-executes the Gaussian process regression algorithm to fit the response surface. The newly fitted response surface function expression replaces the original function expression corresponding to the equipment identification code in the model library.

[0028] It is understandable that the basic vibration data generated by the equipment during operation consists of the superposition of three energy components. The first component is the vibration generated by the operation of the equipment's own mechanical structure. This vibration changes regularly according to the inherent dynamic characteristics of the equipment, depending on the operating parameters such as spindle speed, feed rate, and load current. The second component is the influence of the rigidity characteristics and settlement state of the equipment's installation foundation on the vibration transmission path. A foundation with poor rigidity or large settlement will change the damping characteristics of the contact surface between the equipment base and the foundation, resulting in an overall increase or decrease in the vibration amplitude measured by the sensor under the same operating conditions. The third component is the abnormal vibration component generated when internal components of the equipment experience wear, loosening, cracks, or other faults. If the raw vibration data is used directly to assess the health status, the fixed deviation caused by the foundation factors in the second component will be misjudged as fault characteristics by the artificial intelligence model, causing equipment installed on a slightly soft foundation to be continuously marked as being in a sub-healthy state. By inputting the foundation stiffness coefficient and cumulative foundation settlement as structural parameters into the Gaussian process regression algorithm, the algorithm learns the mapping relationship between the foundation stiffness coefficient and the overall offset of vibration amplitude, as well as the mapping relationship between the cumulative foundation settlement and the drift trend of vibration amplitude over time, during the process of fitting the response surface. Thus, the vibration offset caused by foundation factors is included as a predictable part in the predicted baseline value on the response surface. When the vibration difference value is obtained by subtracting the predicted baseline value from the real-time collected vibration data, the fixed deviation caused by foundation factors has been completely eliminated. The remaining vibration difference value only contains the random disturbances introduced by the fluctuation of the equipment's own operating conditions and the abnormal components generated by the fault.

[0029] The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; In a preferred embodiment of the present invention, calculating the vibration difference value includes: The central processing unit (CPU) sends real-time data read commands to the device controller via industrial Ethernet. Upon receiving the commands, the device controller reads the spindle speed from the spindle driver, the feed rate from the feed servo driver, and the load current from the analog input module. The device controller packages these three values ​​into a data frame and returns it to the CPU via industrial Ethernet. The CPU then parses the spindle speed, feed rate, and load current values ​​from the data frame and uses these three values ​​as query coordinates.

[0030] The central processing unit (CPU) retrieves the corresponding response surface function expression for the device from the model library based on the device identification code. The CPU uses the spindle speed, feed rate, and load current values ​​from the query coordinates as the X, Y, and Z axes in three-dimensional space, respectively, and sets these three values ​​as the independent variables in the response surface function expression. The response surface is stored in a discrete grid format. Grid nodes are divided in 100 rpm intervals along the spindle speed dimension (e.g., 8000 rpm and 8100 rpm each constitute a grid node); in the feed rate dimension, they are divided in 10 mm / min intervals; and in the load current dimension, they are divided in 1 ampere intervals. When the query coordinate value falls exactly on a grid node, the CPU directly reads the predicted baseline value stored at that node. When the value of the queried coordinates falls within a grid cell, the CPU performs trilinear interpolation. Based on the predicted baseline values ​​of the eight vertices of the grid cell containing the queried coordinates, it performs two linear interpolations along the spindle speed dimension to obtain the interpolation results for the two edges, two linear interpolations along the feed rate dimension to obtain the interpolation results for the four faces, and one linear interpolation along the load current dimension to obtain the final predicted baseline value. For example, when the queried coordinates are a spindle speed of 8050 rpm, a feed rate of 505 mm / min, and a load current of 46 amps, the CPU locates a grid cell consisting of a spindle speed of 8000 to 8100 rpm, a feed rate of 500 to 510 mm / min, and a load current of 46 to 47 amps, and calculates a predicted baseline value of 0.18 mm / s.

[0031] The distributed data acquisition terminal continuously acquires vibration acceleration time-domain signals from the triaxial accelerometers at four corner points according to the sampling frequency. The acquired analog signals are converted from analog to digital and stored in the buffer of the distributed data acquisition terminal in time-series form. The central processing unit (CPU) sends a data extraction command to the distributed data acquisition terminal, which then transmits the vibration acceleration time-domain signals in the buffer to the CPU. The CPU performs a Fast Fourier Transform (FFT) on the vibration acceleration time-domain signals at each of the four corner points, converting the time-domain signals into frequency-domain amplitude spectra. It locates the frequency point corresponding to the natural frequency of the device structure in the frequency-domain amplitude spectrum at each corner point and extracts the amplitude at that frequency point. The arithmetic mean of the extracted amplitudes from the four corner points is taken as the real-time baseline vibration data acquired by the device. The CPU subtracts the real-time baseline vibration data from the predicted baseline value. If the real-time baseline vibration data is 0.21 mm / s and the predicted baseline value is 0.18 mm / s, the vibration difference is 0.03 mm / s. The positive or negative sign indicates that the real-time baseline vibration data is higher than the predicted baseline value, and the absolute value of 0.03 mm / s indicates the deviation magnitude.

[0032] Collect environmental interference parameters at the location of each device, and merge the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. In another preferred embodiment of the present invention, separating environmental interference components includes: The central processing unit deploys the regression analyzer in an edge server, which is connected to the IoT gateway via an industrial Ethernet network. The regression analyzer allocates a storage area in the edge server's memory. This storage area uses a key-value database structure, with the combination of device identification code and data collection timestamp as the key, and a vector consisting of vibration difference values, ambient temperature values, ambient humidity values, and operating status parameters of nearby devices as the values. The data records collected from all devices in the workshop during the historical time period are stored in this storage area, forming a historical fusion feature vector set.

[0033] The regression analyzer extracts all records from the historical fusion feature vector set, using the vibration difference value in each record as the dependent variable and the ambient temperature, ambient humidity, and operating status parameters of nearby equipment as independent variables. The ambient temperature is collected by an infrared temperature sensor positioned above the equipment location, which uploads the temperature value (e.g., 25.3 degrees Celsius) to the edge server every 5 minutes. The ambient humidity is collected by a capacitive humidity sensor positioned to the side of the equipment location, which uploads the humidity percentage (e.g., 65%) to the edge server every 5 minutes. The operating status parameters of nearby equipment are obtained by the equipment controller reading the spindle speed, feed rate, and load current of the nearby equipment, then weighting and summing these three values ​​to convert them into a single value (e.g., a weighted sum of the nearby equipment's operating status parameter value of 320).

[0034] The regression analyzer calls the least squares calculation program. The dependent and independent variables are input into the program, which organizes the data into a matrix to construct a design matrix. The number of rows in the design matrix is ​​the total number of historical records, and the number of columns is the number of independent variables plus a constant term column. The program solves the normal equation to obtain the partial regression coefficients for each environmental variable. For example, the partial regression coefficient for ambient temperature is 0.002 mm / s per degree Celsius, the partial regression coefficient for ambient humidity is 0.001 mm / s per percentage point, and the partial regression coefficient for the operating status parameters of nearby equipment is 0.0001 mm / s per unit operating status value.

[0035] The central processing unit (CPU) acquires in real-time vibration differential values, ambient temperature values, ambient humidity values, and nearby device operating status parameters at the current acquisition moment through IoT gateways. The CPU multiplies the ambient temperature value by its corresponding partial regression coefficient (e.g., 25.3 degrees Celsius multiplied by 0.002 mm / s, resulting in 0.0506 mm / s per degree Celsius); it multiplies the ambient humidity value by its corresponding partial regression coefficient (e.g., 65% multiplied by 0.001 mm / s, resulting in 0.065 mm / s per percentage point); and it multiplies the nearby device operating status parameters by their corresponding partial regression coefficients (e.g., 320 multiplied by 0.0001 mm / s, resulting in 0.032 mm / s per unit operating status value). The CPU sums these three products to obtain the environmental disturbance component (e.g., 0.0506 mm / s + 0.065 mm / s + 0.032 mm / s, resulting in 0.1476 mm / s). The CPU subtracts the environmental disturbance component from the vibration differential value to obtain the fault characteristic component.

[0036] The equipment health status assessment results are output based on the magnitude and trend of the fault characteristic components.

[0037] In another preferred embodiment of the present invention, the output device health status assessment result includes: The central processing unit (CPU) extracts all fault feature components corresponding to the current device identification code from the storage area of ​​the edge server. These components are then arranged in chronological order according to their associated acquisition timestamps, forming a fault feature component time series. The CPU performs moving window smoothing on this time series, setting the window width to 5 acquisition cycles. For example, with an acquisition cycle of 10 minutes, the window covers the values ​​of the 5 fault feature components within the most recent 50 minutes. The CPU calculates the arithmetic mean of the 5 values ​​within the window and uses this average as the smoothed fault feature component value for the current moment. The window is then moved forward by one acquisition cycle, and the next average is calculated. This process is repeated until the entire time series is processed, resulting in the smoothed fault feature component time series.

[0038] The central processing unit (CPU) reads the warning threshold corresponding to the device identification code from the operation and maintenance management database. The warning threshold is a pre-set fixed value, for example, 0.5 mm / s. The CPU extracts the last value in the smoothed fault feature component time series as the current value and compares the current value with the warning threshold. If the current value is 0.52 mm / s and the warning threshold is 0.5 mm / s, then the current value exceeds the warning threshold. After determining that the current value exceeds the warning threshold, the CPU extracts the values ​​of the most recent 10 acquisition cycles in the smoothed fault feature component time series. For example, if the most recent 10 values ​​are 0.32 mm / s, 0.35 mm / s, 0.38 mm / s, 0.41 mm / s, 0.44 mm / s, 0.47 mm / s, 0.49 mm / s, 0.51 mm / s, 0.52 mm / s, and 0.52 mm / s, the CPU performs linear regression analysis on these 10 values ​​and calculates the rate of change of the slope of the fitted line. If the slope change rate is positive and shows an increasing trend over three consecutive acquisition cycles, for example, the slope change rate increases from 0.01 mm / s / cycle to 0.03 mm / s / cycle and then to 0.05 mm / s / cycle, the central processing unit outputs a level one warning indicator; if the slope change rate is positive but shows a decreasing trend, for example, the slope change rate decreases from 0.05 mm / s / cycle to 0.03 mm / s / cycle and then to 0.01 mm / s / cycle, the central processing unit outputs a level two warning indicator.

[0039] The central processing unit (CPU) packages the equipment health status assessment results, equipment unique identifier, location identifier, and fault characteristic component time series into a single data packet. The data packet format uses JavaScript object abbreviation format. The CPU transmits the data packet to the operation and maintenance management database via industrial Ethernet. The operation and maintenance management database parses the data packet and stores it in the corresponding data table. The data table uses the equipment unique identifier as the primary key and stores the equipment health status assessment results, location identifier, and fault characteristic component time series as field values.

[0040] This invention addresses the technical problem in intelligent manufacturing where uneven foundation settlement and rigidity differences at different locations within the same workshop lead to fixed deviations in the vibration baseline of equipment of the same model. This causes artificial intelligence models to misjudge equipment with slightly softer foundations and higher vibrations as having extremely poor health and prioritize maintenance. The invention proposes a layered, peeling-off operation and maintenance management method. First, by collecting the location identification code of the equipment, the invention retrieves the foundation rigidity coefficient and cumulative foundation settlement at the corresponding location from the foundation monitoring database. An independent dynamic baseline model is established for each piece of equipment. This model uses the foundation rigidity coefficient and cumulative foundation settlement as structural parameters, and operating condition parameters such as spindle speed, feed rate, and load current as dynamic inputs. It fits a response surface of the equipment foundation vibration data as the operating condition parameters change. The real-time collected foundation vibration data is then differentially calculated with the predicted baseline value output from the response surface to obtain the vibration difference value. This process separates the fixed vibration deviation caused by the physical differences in the foundation from the original data. Subsequently, the method collects ambient temperature, humidity, and operating status parameters of nearby equipment at the equipment location as environmental interference parameters. The vibration difference value and environmental interference parameters are input into a regression analyzer, and the partial regression coefficients of each environmental variable are solved using a multiple linear regression equation. The environmental interference component introduced by the environmental interference parameters is separated from the vibration difference value, yielding the fault characteristic component. This method solves the aforementioned technical problem because it recognizes that equipment vibration data is actually composed of the superposition of four energy components: foundation physical factors, operating condition factors, dynamic environmental factors, and equipment fault factors. The fixed deviation caused by foundation physical factors is strongly correlated with equipment location but weakly correlated with time. Vibration changes caused by operating condition factors can be predicted through nonlinear regression modeling. Fluctuations caused by dynamic environmental factors are linearly correlated with temperature, humidity, and the status of nearby equipment, while the abnormal component generated by equipment fault factors is unrelated to the first three. By embedding the foundation rigidity coefficient and cumulative foundation settlement as structural parameters into the nonlinear regression model, the model automatically incorporates the vibration offset caused by foundation factors into the predictable part during the learning process, ensuring that the vibration difference value after differential calculation no longer includes the fixed foundation deviation. Then, by using a regression analyzer to learn the linear relationship between environmental disturbance parameters and vibration difference values ​​using historical data, the disturbance components introduced by dynamic environmental factors are further stripped away. The final fault characteristic components purely reflect the fault states of the equipment's own components, such as wear, loosening, and cracks. This avoids misjudgments caused by foundation differences and environmental interference, and ensures that the equipment health status assessment results truly reflect the health status of the equipment itself.

[0041] An IoT-based AI-powered operation and maintenance management system includes: Initial module: Collects basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; Prediction module: The equipment dynamic baseline model is established for each piece of equipment using the equipment installation environment feature set as input. The equipment dynamic baseline model outputs the predicted baseline value of the foundation vibration data. The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; Optimization module: Collects environmental interference parameters at the location of each device, and merges the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. Management module: Outputs equipment health status assessment results based on the magnitude and trend of fault characteristic components.

[0042] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. An AI-based operation and maintenance management method based on the Internet of Things, characterized in that, Includes the following steps: Collect basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; A dynamic baseline model for each piece of equipment is established using the equipment installation environment feature set as input. The dynamic baseline model outputs the predicted baseline value of the foundation vibration data. The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; Collect environmental interference parameters at the location of each device, and merge the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. The equipment health status assessment results are output based on the magnitude and trend of the fault characteristic components.

2. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, The data collected includes basic vibration data, equipment operating parameters, and location identification codes for each piece of equipment in the production workshop. During equipment installation, triaxial accelerometers are placed at the four corners of the contact surface between the equipment base and the foundation. The triaxial accelerometers are synchronized with the clock of the central control system in the production workshop. The triaxial accelerometers continuously collect the vibration acceleration time-domain signal of the equipment under no-load operation. The vibration acceleration time-domain signal is Fourier transformed to obtain the frequency domain amplitude spectrum. The amplitude corresponding to the natural frequency of the equipment structure is extracted from the frequency domain amplitude spectrum and recorded as the basic vibration data. The equipment controller reads the spindle speed, feed speed, and load current of the equipment in real time and records them as operating parameters. The coordinate code stored in the RFID tag embedded in the foundation is used as the location identification code of the equipment, and the reading range of the RFID tag is limited to the coverage area of ​​the equipment installation base.

3. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, Extracting foundation response characteristic parameters includes: The foundation monitoring database pre-stores the foundation rigidity coefficient and cumulative foundation settlement corresponding to each location identifier. The IoT gateway retrieves the foundation rigidity coefficient and cumulative foundation settlement from the foundation monitoring database using the location identifier as an index and records them as foundation response characteristic parameters. The foundation rigidity coefficient is obtained by conducting an impact excitation test on the foundation before equipment installation. During the impact excitation test, a standard pulse load is applied to the center point of the equipment installation location, and a displacement sensor is placed at a preset distance from the center point. The peak displacement response is recorded by the displacement sensor, and the ratio of the amplitude of the standard pulse load to the peak displacement response is defined as the foundation rigidity coefficient. The cumulative foundation settlement is obtained by continuously collecting data using a static level instrument placed around the equipment installation location. The elevation change value is recorded every 24 hours by the static level instrument, and the difference between the elevation change value and the initial installation elevation is accumulated to obtain the cumulative foundation settlement.

4. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, Obtaining the predicted baseline value includes: The foundation rigidity coefficient and cumulative foundation settlement, which are part of the equipment installation environment feature set, are used as fixed inputs to the model's structural parameters. The equipment operating condition parameters are used as dynamic inputs to the model, and the equipment foundation vibration data are used as the model's output target. A nonlinear regression algorithm is used to fit the response surface of the foundation vibration data as the equipment operating condition parameters change within the operating condition parameter space. This response surface is recorded as the equipment's dynamic baseline model, with each operating condition parameter point on the response surface corresponding to a predicted baseline value. The equipment's unique identification code is associated with and stored in relation to the function expression of the response surface. Each time a new cumulative foundation settlement is obtained through a hydrostatic level, the nonlinear regression algorithm is re-executed to fit the response surface.

5. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, The calculation of vibration differential values ​​includes: The equipment controller reads the spindle speed, feed rate, and load current in real time and inputs these values ​​as query coordinates into the equipment dynamic baseline model. The dynamic baseline model then performs trilinear interpolation on the response surface based on the query coordinates to obtain the predicted baseline value. A triaxial accelerometer collects the vibration acceleration time-domain signal in real time and converts it into an effective vibration value. The effective vibration value is then subtracted from the predicted baseline value to obtain the vibration difference value. The sign of the vibration difference value indicates the direction of deviation of the effective vibration value from the predicted baseline value, and the absolute value of the vibration difference value indicates the magnitude of the deviation.

6. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, Separating environmental interference components includes: The regression analyzer pre-stores a set of historical fusion feature vectors for all equipment in the workshop. This set includes vibration difference values ​​and corresponding ambient temperature, humidity, and operating status parameters of nearby equipment. The ambient temperature, humidity, and operating status parameters are denoted as environmental disturbance parameters. The regression analyzer establishes a multiple linear regression equation with the vibration difference value as the dependent variable and the ambient temperature, humidity, and operating status parameters as independent variables. It then uses the least squares method to solve for the partial regression coefficients of each environmental variable in the multiple linear regression equation. The environmental disturbance component is obtained by summing the products of the ambient temperature value and its corresponding partial regression coefficient, the ambient humidity value and its corresponding partial regression coefficient, and the operating status parameters of nearby equipment and their corresponding partial regression coefficients. The fault characteristic component is obtained by subtracting the environmental disturbance component from the vibration difference value.

7. The IoT-based AI-powered operation and maintenance management method according to claim 1, characterized in that, The output device health status assessment results include: The fault feature components are arranged in order of their acquisition timestamps to form a fault feature component time series. A moving window smoothing process is then applied to this time series to obtain a smoothed fault feature component time series. The current value of the smoothed fault feature component time series is compared with a preset warning threshold. When the current value exceeds the warning threshold, the slope change rate of the smoothed fault feature component time series over the past continuous acquisition periods is extracted. If the slope change rate shows an increasing trend, a first-level warning indicator is output; if the slope change rate shows a decreasing trend, a second-level warning indicator is output. The equipment health status assessment results, the equipment unique identifier, the location identifier, and the fault feature component time series are packaged and stored in the operation and maintenance management database.

8. An AI-based operation and maintenance management system based on the Internet of Things, characterized in that, include: Initial module: Collects basic vibration data, equipment operating parameters, and location identification codes of each piece of equipment in the production workshop; The IoT gateway retrieves the foundation response characteristic parameters of the corresponding location based on the location identification code, and merges the foundation response characteristic parameters with the foundation vibration data into a set of equipment installation environment features; Prediction module: The equipment dynamic baseline model is established for each piece of equipment using the equipment installation environment feature set as input. The equipment dynamic baseline model outputs the predicted baseline value of the foundation vibration data. The vibration difference value is obtained by performing a difference operation between the collected basic vibration data and the predicted baseline value; Optimization module: Collects environmental interference parameters at the location of each device, and merges the environmental interference parameters with vibration differential values, equipment operating condition parameters, and location identification codes into a fusion feature vector; The fused feature vector is input into the regression analyzer to separate the environmental disturbance component introduced by the environmental disturbance parameter from the vibration difference value, and the remaining part is defined as the fault feature component. Management module: Outputs equipment health status assessment results based on the magnitude and trend of fault characteristic components.

Images