A method and system for predicting the health of industrial equipment based on multi-source data synchronization

Abstract

The present invention relates to the technical field of predictive maintenance of industrial equipment, and discloses a method and system for predicting the health of industrial equipment based on multi-source data synchronization. The method comprises: synchronously collecting vibration spectra, temperature gradients, and current harmonics in response to MOM system instructions; reorganizing the vibration spectrum based on spindle period slicing; calculating the cross-correlation delay offset between current harmonics and temperature gradients and translating the temperature data; integrating multi-source data according to the spindle slicing time window to generate a synchronization signal matrix; inputting a wear degradation model to generate a health decline index; and triggering a maintenance response when the index exceeds the dynamic threshold of the MOM system. The present invention solves the problems of asynchronous fusion distortion of multi-source data and insufficient adaptability of static strategies through vibration phase synchronization, delay compensation, and dynamic threshold decision-making, thereby improving the accuracy of equipment status prediction and maintenance efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of predictive maintenance of industrial equipment, and in particular to a method and system for predicting the health of industrial equipment based on multi-source data synchronization. Background Art

[0002] In process manufacturing fields such as rubber and tires, the health of industrial equipment directly affects production continuity and product quality. Currently, companies generally use MOM (Manufacturing Operations Management) systems to coordinate production plans, but equipment health prediction still relies on a single data source or asynchronous multi-sensor monitoring, which presents significant bottlenecks:

[0003] Key parameters such as vibration spectra, temperature gradients, and current harmonics have inherent timing deviations due to differences in physical transmission characteristics and sampling mechanisms. For example, in the early stages of spindle bearing wear in rubber equipment, the vibration characteristics must be strictly synchronized with the spindle's rotation phase to be identified. When vibration analysis is separated from the equipment's spindle phase cycle, it is unable to capture periodic wear characteristics. Furthermore, the lag between temperature response and current changes is not compensated. Direct fusion of asynchronous data can overwhelm early equipment fault characteristics with noise, resulting in missed detections.

[0004] The MOM system's maintenance strategy library often presets fixed health thresholds, making it unable to adapt to the dynamic changes in process parameters such as temperature and pressure during the rubber vulcanization process. During the high-temperature vulcanization stage, normal thermal expansion of the equipment causes abnormal fluctuations in the health index, frequently triggering false maintenance work orders and disrupting production scheduling. Furthermore, when key components actually degrade, the rigid thresholds lead to delayed responses. These issues severely limit the large-scale application of predictive maintenance in the rubber industry. Summary of the Invention

[0005] The present invention provides an industrial equipment health prediction method and system based on multi-source data synchronization, the main purpose of which is to solve the problems of equipment status misjudgment and maintenance response inaccuracy caused by the spatiotemporal asynchrony of vibration / temperature / current data and the rigidity of static health threshold decision-making.

[0006] To achieve the above objectives, the present invention provides an industrial equipment health prediction method based on multi-source data synchronization, comprising:

[0007] S1, responds to the data collection instructions issued by the MOM system based on the production plan and real-time status of the equipment, and synchronously collects the vibration spectrum, temperature gradient and current harmonics of the rubber equipment;

[0008] S2. Slicing and recombining the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase;

[0009] S3. Calculate the cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset;

[0010] S4. performing a time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient;

[0011] S5, integrating the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window;

[0012] S6. Inputting the synchronization signal matrix into a wear degradation model of a rubber device to generate a health degradation index of the rubber device;

[0013] S7. When the health decline index exceeds the health threshold defined by the MOM system maintenance policy library, a device maintenance work order is generated and the MOM system is triggered to execute a maintenance response.

[0014] Optionally, the synchronous collection of the vibration spectrum, temperature gradient and current harmonics of the rubber equipment includes:

[0015] The vibration spectrum of the rubber equipment is collected by an acceleration sensor;

[0016] collecting the temperature gradient of the rubber device by an infrared thermal imager array;

[0017] The current harmonics of the rubber device are collected by Hall effect sensors, wherein the sensor group is deployed on the main shaft bearing seat of the rubber device.

[0018] Optionally, slicing and reorganizing the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase includes:

[0019] Determine the vibration period division point based on the pulse signal output by the spindle encoder;

[0020] Based on the vibration period division point, the vibration spectrum is sliced ​​at equal phase angles, wherein the time domain width of each phase slice is less than one twelfth of the complete rotation period of the main shaft;

[0021] The Hanning window function is used to intercept the time domain waveform corresponding to the equal phase angle slice;

[0022] The intercepted time domain waveform segments are converted into spectrum slices by fast Fourier transform, and the spectrum slices are collected as a vibration slice sequence synchronized with the main axis phase, wherein the expression of the fast Fourier transform is:

[0023] ;

[0024] in, No. The complex value of the frequency point, The bearing seat vibration time domain signal The amplitude of the sampling points, is the time domain sampling point index, is the spectrum index number, is the number of sampling points in a single cycle, Is an imaginary unit.

[0025] Optionally, calculating the cross-correlation function between the current harmonics and the temperature gradient to determine the time delay offset includes:

[0026] Converting the current harmonics into a time domain envelope signal;

[0027] Calculating the Pearson correlation coefficient between the time domain envelope signal and the temperature gradient through a sliding window;

[0028] The peak position of the Pearson correlation coefficient is taken as the time delay offset.

[0029] Optionally, the calculation formula of the Pearson correlation coefficient is:

[0030] ;

[0031] in, is the Pearson correlation coefficient, The temperature gradient Sampling point values, is the mean temperature gradient within the window, is the current envelope Sampling point values, is the mean value of the current envelope in the window, is the thermal conductivity response window length of the rubber device, is the sliding window sampling point index.

[0032] Optionally, performing a time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient includes:

[0033] Determining a time coordinate offset value of the temperature gradient based on the time delay offset;

[0034] Linear interpolation compensation is performed on the temperature gradient to generate a phase-synchronized temperature gradient signal, wherein a temperature gradient sampling moment is aligned with a current harmonic sampling moment.

[0035] Optionally, the step of integrating the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window includes:

[0036] Extracting energy entropy values ​​of vibration slice features in the vibration slice sequence by wavelet packet decomposition, and using the energy entropy values ​​as first dimensional features;

[0037] Calculating the mean value of the temperature gradient after phase synchronization within the main axis slice time window as a second dimension feature;

[0038] Extracting the total harmonic distortion rate of the current harmonics as a third dimension feature;

[0039] According to the time sequence of the main axis slice time window, the first dimensional features, the second dimensional features and the third dimensional features are combined into a three-dimensional synchronization signal matrix.

[0040] Optionally, the energy entropy value of the vibration slice feature in the vibration slice sequence is extracted by wavelet packet decomposition, wherein the energy entropy value is calculated as follows:

[0041] ;

[0042] in, is the energy entropy that characterizes the degree of wear, is the total number of characteristic frequency band coefficients of rubber equipment bearings, It is Node wavelet packet coefficients, is the wavelet packet coefficient index, is the total number of wavelet packet coefficients, is the wavelet decomposition node index.

[0043] Optionally, the wear degradation model is a convolutional neural network model, which is obtained by training equipment maintenance work order data corresponding to vibration spectrum anomalies, temperature gradient mutations and current harmonic distortion, wherein the input layer dimension of the wear degradation model matches the three-dimensional feature structure of the synchronization signal matrix, and the output layer is a fully connected layer and is mapped to a scalar value of the health decline index.

[0044] In order to solve the above problems, the present invention further provides an industrial equipment health prediction system based on multi-source data synchronization, the system comprising:

[0045] The multi-source data acquisition module is used to respond to data acquisition instructions issued by the MOM system based on production plans and real-time equipment status, and synchronously collect vibration spectra, temperature gradients, and current harmonics of rubber equipment;

[0046] A vibration spectrum slicing and reorganization module is used to slice and reorganize the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase;

[0047] a time delay offset determination module, configured to calculate a cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset;

[0048] A temperature gradient time domain shift module, configured to perform time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient;

[0049] a synchronization signal matrix generation module, configured to integrate the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to a main axis slicing time window;

[0050] a health decay index generating module, configured to input the synchronization signal matrix into a wear degradation model of a rubber device to generate a health decay index of the rubber device;

[0051] The equipment maintenance work order generation module is used to generate an equipment maintenance work order and trigger the MOM system to execute a maintenance response when the health decline index exceeds the health threshold defined by the MOM system maintenance policy library.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. By reorganizing vibration spectrum slices based on the spindle's operating cycle, strict synchronization of vibration characteristics and the equipment's rotation phase is achieved, making periodic wear characteristics identifiable. Furthermore, by compensating for the cross-correlation delay between current harmonics and temperature gradients, cross-physical field data transmission delays are eliminated, ensuring the spatiotemporal consistency of multi-source feature fusion. As a result, the missed detection rate of early-stage hidden wear of rubber equipment is significantly reduced, and the ability to identify complex faults is improved.

[0054] 2. The synchronization signal matrix is ​​input into the convolutional neural network model to output a quantitative health decline index, breaking through the limitations of single-parameter analysis. The health threshold is dynamically adjusted based on the MOM system maintenance strategy library to adapt to the variable operating conditions of the rubber vulcanization process and avoid false triggering caused by interference such as normal thermal expansion. While ensuring prediction accuracy, it reduces unnecessary maintenance work orders and achieves efficient scheduling of production resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of an industrial equipment health prediction method based on multi-source data synchronization provided by one embodiment of the present invention;

[0056] Figure 2 This is a functional module diagram of an industrial equipment health prediction system based on multi-source data synchronization provided by one embodiment of the present invention;

[0057] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0058] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] An embodiment of the present application provides an industrial equipment health prediction method based on multi-source data synchronization. The execution subject of the industrial equipment health prediction method based on multi-source data synchronization includes but is not limited to at least one of the electronic devices such as the server, terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the industrial equipment health prediction method based on multi-source data synchronization can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks, and big data and artificial intelligence platforms.

[0060] Reference Figure 1 FIG2 is a flow chart of an industrial equipment health prediction method based on multi-source data synchronization provided by an embodiment of the present invention. In this embodiment, the industrial equipment health prediction method based on multi-source data synchronization includes:

[0061] S1. Respond to the data collection instructions issued by the MOM system based on the production plan and real-time status of the equipment, and synchronously collect the vibration spectrum, temperature gradient and current harmonics of the rubber equipment.

[0062] In an embodiment of the present invention, the synchronous collection of the vibration spectrum, temperature gradient and current harmonics of the rubber equipment includes:

[0063] The vibration spectrum of the rubber equipment is collected by an acceleration sensor;

[0064] collecting the temperature gradient of the rubber device by an infrared thermal imager array;

[0065] The current harmonics of the rubber device are collected by Hall effect sensors, wherein the sensor group is deployed on the main shaft bearing seat of the rubber device.

[0066] Specifically, in the rubber equipment production process, existing technologies primarily rely on single-sensor data or asynchronous multi-source data for equipment health prediction, resulting in data that cannot fully reflect the actual operating status of the equipment. For example, when relying solely on vibration spectrum analysis, the correlation between temperature anomalies and current fluctuations cannot be combined, and hidden equipment wear is often missed, resulting in increased unplanned equipment downtime and increased maintenance costs. Furthermore, traditional data collection methods lack real-time linkage with the MOM system, and data collection is not timely and targeted enough, making it difficult to meet the rubber vulcanization process's demand for real-time monitoring of equipment status.

[0067] Specifically, the MOM system (manufacturing operations management system) can integrate production plans with real-time equipment status data, and has functions such as command issuance, data processing, and maintenance response, enabling operational management of the entire life cycle of rubber equipment.

[0068] Specifically, the vibration spectrum refers to the characteristic data generated by the vibration signal generated by rubber equipment during operation. This data is collected by an accelerometer and converted into the frequency domain. This data contains the frequency components and energy distribution of the vibrations in different parts of the equipment. By analyzing this data, it can be determined whether the equipment is experiencing abnormal vibrations.

[0069] Specifically, the temperature gradient is the spatial distribution of surface temperature differences in rubber equipment, captured by an infrared thermal imaging array. During the rubber vulcanization process, temperature variations across the equipment can reflect its operating status. Issues such as bearing wear and motor overload can cause abnormal temperature fluctuations.

[0070] Specifically, an accelerometer is a sensor mounted on the main shaft bearing of a rubber equipment. Its function is to convert the acceleration generated by the equipment's vibration into an electrical signal, thereby enabling the acquisition of the vibration spectrum. The accelerometer selected in this invention has high sensitivity and a wide frequency response range, capable of accurately capturing weak vibrations of the equipment.

[0071] Specifically, the infrared thermal imaging array, consisting of multiple infrared cameras deployed around the main shaft bearing housing of the rubber equipment, monitors the equipment surface temperature in real time, generating temperature gradient data. The array boasts high temperature measurement accuracy and spatial resolution, enabling timely detection of abnormal temperature areas within the equipment.

[0072] Specifically, a Hall Effect sensor utilizes the Hall Effect principle and is connected in series to the power supply circuit of the spindle motor of a rubber equipment to collect current harmonics. It detects the magnetic field changes generated by the current, thereby capturing the harmonic components in the current and providing a basis for analyzing the electrical status of the equipment.

[0073] Specifically, considering that the main shaft bearing seat of the rubber equipment is a key location for vibration, heat generation, and current fluctuations, an accelerometer, an infrared thermal imager array, and a Hall effect sensor are deployed there. The accelerometer can capture vibrations at various speeds; the infrared thermal imager array, consisting of eight IR-T600 thermal imagers, can capture subtle changes in the equipment's surface temperature in real time; and the Hall effect sensor, a closed-loop HL-2000 sensor, accurately captures the spindle motor's current harmonics.

[0074] Specifically, the data acquisition terminal features multi-channel synchronous data acquisition, simultaneously receiving data from three sensors and performing preliminary data pre-processing such as filtering and amplification. The data acquisition terminal is connected to the MOM system via industrial Ethernet, ensuring real-time and stable data transmission.

[0075] Furthermore, the synchronous acquisition control algorithm refers to the MOM system generating data acquisition instructions based on the production plan and the real-time status of the equipment. The instructions include parameters such as the acquisition time interval and duration. After receiving the instructions, the data acquisition terminal triggers the sensor group through the internal synchronous clock to begin data acquisition, ensuring the consistent acquisition timing of the vibration spectrum, temperature gradient, and current harmonics. For example, when the MOM system detects that the rubber vulcanization process has entered a certain stage, it will issue an acquisition instruction with an interval of 100ms. After receiving the instruction, the data acquisition terminal will simultaneously activate the three sensors to collect data, with each acquisition lasting 500ms.

[0076] Specifically, in the rubber equipment health prediction scenario, the initial data are set as follows: spindle speed 1500 r / min, sampling frequency 20 kHz.

[0077] Furthermore, the fundamental frequency can be calculated from the spindle speed of 1500 rpm to be 25 Hz (1500 ÷ 60 = 25). According to the Nyquist sampling theorem, the sampling frequency is set to 20 kHz, which meets the requirement of being at least twice the highest frequency component. In step S2, the vibration period is calculated based on the fundamental frequency to be 0.04 s (1 ÷ 25 = 0.04), thereby determining the vibration period division point.

[0078] In general, the MOM system is deeply integrated with multi-sensor data collection for the first time, realizing dynamic data collection based on production plans and real-time status of equipment, and improving the pertinence and timeliness of data collection.

[0079] In general, deploying multiple sensors on the spindle bearing housing enables the simultaneous collection of multiple data sources, including vibration, temperature, and current, resolving the analytical errors caused by asynchronous data collection in existing technologies. For example, during the rubber vulcanization process, bearing wear can simultaneously cause abnormal vibration, elevated temperatures, and current fluctuations. By synchronously collecting these three types of data, the equipment's status can be more accurately determined.

[0080] Specifically, when the MOM system detects that the rubber equipment's current production schedule is to vulcanize a certain tire model and that the equipment's real-time status indicates a spindle speed of 1500 rpm, it issues a data collection command. The accelerometer collects vibration signals at a sampling frequency of 20 kHz, the infrared thermal imaging array collects temperature gradient data at a sampling frequency of 30 Hz, and the Hall effect sensor collects current harmonics at a sampling frequency of 20 kHz.

[0081] For example, at a certain acquisition time t0, the amplitude of a certain sampling point in the vibration time domain signal collected by the acceleration sensor is 0.5g; the infrared thermal imager array detects that the temperature of a certain area of ​​the spindle bearing seat is 65°C; the current value collected by the Hall effect sensor is 120A, of which the 5th harmonic component is 12A.

[0082] The output of step S1 is the raw data for the vibration spectrum, temperature gradient, and current harmonics. This data serves as input for the subsequent vibration spectrum slicing and reorganization in step S2 and also forms the basis for calculating the cross-correlation function between current harmonics and temperature gradients in step S3. For example, the raw vibration spectrum data collected in step S1 will be sliced ​​and reorganized based on the spindle operating cycle in step S2, while the collected current harmonics and temperature gradient data will be used in step S3 to calculate the time delay offset to achieve data time synchronization.

[0083] S2. Slice and reorganize the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence that is synchronized with the spindle phase.

[0084] In an embodiment of the present invention, slicing and reorganizing the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase includes:

[0085] Determine the vibration period division point based on the pulse signal output by the spindle encoder;

[0086] Based on the vibration period division point, the vibration spectrum is sliced ​​at equal phase angles, wherein the time domain width of each phase slice is less than one twelfth of the complete rotation period of the main shaft;

[0087] The Hanning window function is used to intercept the time domain waveform corresponding to the equal phase angle slice;

[0088] The intercepted time domain waveform segments are converted into spectrum slices by fast Fourier transform, and the spectrum slices are collected as a vibration slice sequence synchronized with the main axis phase, wherein the expression of the fast Fourier transform is:

[0089] ;

[0090] in, No. The complex value of the frequency point, The bearing seat vibration time domain signal The amplitude of the sampling points, is the time domain sampling point index, is the spectrum index number, is the number of sampling points in a single cycle, Is an imaginary unit.

[0091] During the operation of rubber equipment, existing technologies for processing vibration spectra fail to incorporate spindle phase information, resulting in a disconnect between the vibration characteristics and the actual operating status of the equipment. For example, when traditional methods directly perform spectrum analysis on vibration signals, they are unable to accurately distinguish the vibration differences of the equipment at different spindle rotation phases. This causes characteristic frequencies such as rolling bearing outer ring faults and inner ring faults to be submerged in the overall vibration signal, resulting in the missed detection of hidden equipment damage. Furthermore, vibration data from unsynchronized spindle phases is difficult to establish spatiotemporal correlations with other parameters such as temperature and current, affecting the accuracy of multi-source data fusion and failing to meet the MOM system's requirements for refined health management of rubber equipment.

[0092] Specifically, a spindle encoder is an incremental photoelectric encoder installed on the main shaft of a rubber machine. Customized to the specific characteristics of the machine, it boasts a resolution of 1024 lines per revolution. It measures the spindle's rotation angle and position by outputting pulse signals. Each pulse corresponds to a specific spindle rotation angle, providing a precise time reference for determining the vibration cycle.

[0093] Specifically, the vibration cycle division points are key points used to divide the vibration signal cycle, determined based on the pulse signal output by the spindle encoder. Specifically, the rising or falling edge of the spindle encoder pulse signal is identified as the starting or ending point of a vibration cycle, thereby defining the period range of the vibration signal.

[0094] Specifically, equal phase angle slicing divides the 360° rotation of the main shaft into a number of equal phase intervals. In this invention, to ensure that each phase slice corresponds to the same main shaft phase and improve the accuracy of vibration signature analysis, the main shaft rotation is divided into 12 equal phase intervals of 30° each, thus achieving equal phase angle slicing of the vibration spectrum.

[0095] An SR-PE1024 incremental photoelectric encoder was selected and installed on the non-load end of the rubber equipment spindle. It is directly connected to the spindle via an elastic coupling, ensuring that the encoder accurately reflects the spindle's rotational state. This encoder has a resolution of 1024 lines per revolution, meaning it outputs 1024 pulses for every spindle rotation. A high-speed counter counts these pulses to accurately determine the spindle's rotational angle and position.

[0096] An industrial data acquisition card is used to synchronously collect the pulse signal from the spindle encoder and the vibration signal output by the accelerometer. The acquisition card is connected to the industrial control computer via a PCIe interface to ensure real-time and stable data transmission.

[0097] Based on the pulse signal output by the spindle encoder, a software algorithm determines the vibration cycle division points. Specifically, when a rising edge of the pulse signal is detected, an interrupt event is triggered, and the system records the current time as the starting point of the vibration cycle. To improve the accuracy of cycle division, a dual trigger mechanism is used. That is, a valid vibration cycle starting point is determined only when two rising edge pulse signals are detected consecutively, avoiding cycle division errors caused by interference signals.

[0098] The equal phase angle segmentation algorithm divides the 360° rotation of the main shaft into 12 equal phase intervals, each 30°, based on the vibration period division points. The number of time-domain sampling points corresponding to each phase interval is calculated based on the main shaft rotation speed and sampling frequency, thus achieving equal phase angle segmentation of the vibration spectrum.

[0099] For example, when the spindle speed is 1500r / min, the base frequency is 25Hz, the period is 0.04s, and the sampling frequency is 20kHz, the number of sampling points in each period is The period is divided into 12 phase intervals, and the time domain width corresponding to each phase interval is , the corresponding number of sampling points is That is, the time domain width of each phase slice contains about 66 sampling points, which is less than one twelfth of the complete rotation period of the main shaft.

[0100] The Hanning window truncation algorithm uses a Hanning window function to truncate each time-domain waveform segment after equal phase angle slicing. The length of the window function is equal to the number of time-domain sampling points in each phase slice. By multiplying the time-domain waveform by the Hanning window function, the waveform is smoothly truncated, reducing spectral leakage.

[0101] The Fast Fourier Transform algorithm is to perform a Fast Fourier Transform on the intercepted time domain waveform fragment to convert the time domain signal into a frequency domain spectrum. In the present invention, the number of points P of FFT is 1024, which is 2 to the power of 10, to improve the calculation efficiency. The complex value of each frequency point is obtained by FFT calculation. , where the real part and imaginary part represent the amplitude of the cosine and sine components of the frequency point respectively, and then the amplitude and phase of each frequency point are calculated to obtain the spectrum slice.

[0102] In the rubber equipment health prediction scenario, the initial data are set as follows: spindle speed 1500 r / min, sampling frequency 20 kHz, and spindle encoder resolution 1024 lines / rev.

[0103] The fundamental frequency is calculated from the spindle speed of 1500r / min , according to the sampling frequency of 20kHz, the number of sampling points in a single cycle can be obtained However, in order to facilitate FFT calculation and improve calculation efficiency, P is set to 1024 (2 to the power of 10). In FFT calculation, the spectrum index number The value range is 0 to 1023, where Corresponding to the DC component, Corresponding to the base frequency 25Hz, The corresponding harmonic is 50Hz, and so on.

[0104] The present invention proposes a method for accurately dividing the vibration period based on the spindle encoder pulse signal, achieving strict synchronization between the vibration spectrum and the spindle phase. Conventional techniques often use the vibration signal's autocorrelation function or peak detection to determine the period division, which is susceptible to noise and changes in the device's operating state, leading to inaccurate period division. However, the present invention utilizes the precise position signal provided by the spindle encoder to accurately determine the vibration period division points, aligning each vibration slice with a specific spindle phase and improving the matching of vibration characteristics with the device's actual operating state.

[0105] Furthermore, the accuracy of spectrum analysis is improved by combining equal phase angle slicing with a Hanning window function. Conventional techniques typically slice vibration signals at equal time intervals, which cannot guarantee that each slice corresponds to the same main axis phase, leading to phase ambiguity during spectrum analysis. However, the present invention uses equal phase angle slicing to ensure that each slice corresponds to the same main axis phase. Combined with a Hanning window function, this method reduces spectrum leakage, enabling more accurate extraction of the device's vibration characteristic frequencies, such as the fault characteristic frequencies of rolling bearings.

[0106] Furthermore, the fast Fourier transform is applied to the spectrum analysis of equal phase angle slices, which improves the computational efficiency and the real-time performance of feature extraction. In traditional technology, the spectrum analysis of vibration signals usually adopts discrete Fourier transform (DFT), which has a large amount of computation and is difficult to meet the needs of real-time monitoring. However, the present invention adopts FFT algorithm to reduce the computational complexity from Reduce to , which greatly improves the computational efficiency and enables real-time extraction and analysis of vibration characteristics of rubber equipment.

[0107] When the main shaft speed of the rubber equipment is 1500r / min, the main shaft encoder outputs a pulse signal of 1024 lines / revolution, and the high-speed counter records the rising edge of the pulse signal as the vibration cycle division point. According to the sampling frequency of 20kHz, the number of sampling points in each vibration cycle is theoretically 800, but for the convenience of FFT calculation, Based on the vibration period division point, the main shaft is divided into 12 phase intervals for one rotation. Each phase interval is 30°, and the corresponding time domain width is , the number of sampling points in each phase interval is approximately The Hanning window function is used to intercept the 66 sampling points in each phase interval, and the window function length is , the Hanning window function expression is ,in From 0 to 65. Perform 1024-point FFT transformation on the intercepted time domain waveform segment to obtain spectrum slice. For example, for the time domain sampling points within a certain phase interval , when performing FFT calculation, for (base frequency 25Hz), Assumptions (the rest of the sampling points are 0), then , its modulus represents the amplitude of the fundamental frequency component. By calculation, the amplitude of the fundamental frequency component in this phase interval can be obtained as The spectrum slices of the 12 phase intervals are collected in phase order to form a vibration slice sequence that is synchronized with the main axis phase.

[0108] This step, through the reorganization of the vibration spectrum slices, generates a sequence of vibration slices synchronized with the spindle phase, accurately reflecting the vibration characteristics of the equipment at different spindle phases. Compared with traditional vibration analysis methods that do not synchronize the spindle phase, this method improves the accuracy of identifying the characteristic frequencies of rolling bearing faults, effectively enhancing the ability to detect hidden damage in rubber equipment.

[0109] The inputs to step S2 are the raw time-domain data of the rubber equipment's vibration spectrum collected in step S1 and the pulse signal output by the spindle encoder. By performing period division, equal phase angle slicing, Hanning window truncation, and FFT transformation on the raw time-domain data, a sequence of vibration slices synchronized with the spindle phase is generated. This serves as the first dimension of the feature input for constructing the three-dimensional synchronization signal matrix in the subsequent step S5.

[0110] For example, each spectrum slice in the vibration slice sequence generated by S2 will be fused with the temperature gradient data and current harmonic data that have undergone time domain translation in step S4 to construct a three-dimensional synchronous signal matrix containing vibration, temperature, and current characteristics, providing multi-dimensional feature input for health prediction of the CNN model.

[0111] S3. Calculate the cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset.

[0112] In an embodiment of the present invention, calculating the cross-correlation function between the current harmonics and the temperature gradient to determine the time delay offset includes:

[0113] Converting the current harmonics into a time domain envelope signal;

[0114] Calculating the Pearson correlation coefficient between the time domain envelope signal and the temperature gradient through a sliding window;

[0115] The peak position of the Pearson correlation coefficient is taken as the time delay offset.

[0116] In detail, the calculation formula of the Pearson correlation coefficient is:

[0117] ;

[0118] in, is the Pearson correlation coefficient, The temperature gradient Sampling point values, is the mean temperature gradient within the window, is the current envelope Sampling point values, is the mean value of the current envelope in the window, is the thermal conductivity response window length of the rubber device, is the sliding window sampling point index.

[0119] Specifically, during the operation of rubber equipment, current harmonics and temperature gradients, as important parameters reflecting the equipment's status, have essentially different physical meanings. Current harmonics primarily reflect abnormalities in the equipment's electrical system, while temperature gradients reflect changes in the equipment's thermal state. In existing technologies, because the transmission delay between the two in the time dimension is not considered, directly fusing their features will result in data timing disorder. For example, in the early stages of equipment bearing wear, abnormal changes in current harmonics often precede increases in temperature gradients. Ignoring this time delay will cause the feature vectors after multi-source data fusion to lose their physical meaning, leading to an error of more than 20% in the equipment health status assessment. The MOM system requires real-time health monitoring of rubber equipment, so the time synchronization problem of multi-source data must be solved to ensure the accuracy of the subsequent health prediction model.

[0120] The time-domain envelope signal is the signal obtained by performing envelope detection on the current harmonics. Specifically, the amplitude envelope of the current harmonic signal is extracted through methods such as Hilbert transform or absolute value filtering. This is used to highlight the amplitude variation characteristics of the current signal and facilitate correlation analysis with temperature gradients. In rubber equipment, when the spindle motor experiences a fault such as an interturn short circuit, the amplitude envelope of the current harmonics will change significantly. The time-domain envelope signal can more intuitively reflect this change.

[0121] A sliding window is a fixed-length window that slides across a time series. Its purpose is to calculate the Pearson correlation coefficient segment by segment to capture the changing correlation between the current envelope and the temperature gradient over different time periods. For example, during the rubber vulcanization process, the equipment temperature gradually increases over time, and the current also changes accordingly. Using a sliding window, the correlation between the two can be dynamically analyzed at different vulcanization stages.

[0122] The Pearson correlation coefficient is a statistic that measures the linear correlation between two variables, with a value range of [-1, 1]. In this paper, it is used to measure the degree of linear correlation between the current envelope and the temperature gradient. A larger absolute value indicates a stronger correlation, and the peak position indicates the time offset when the correlation between the two variables is strongest.

[0123] Delay offset refers to the time delay between current harmonics and temperature gradients. Because current changes cause device heating, which requires a certain amount of heat conduction time, there is a time difference between the two. This delay offset is used to subsequently time-shift the temperature gradient to achieve time synchronization between the two.

[0124] Performs hardware envelope detection on the current harmonics collected by the Hall effect sensor to generate a time-domain envelope signal. The module has a bandwidth of 0-10kHz and is able to preserve the low-frequency envelope characteristics of the current harmonics.

[0125] The Hilbert transform method is used to extract the envelope of current harmonics. The specific steps are: Perform Hilbert transform to get , then the time domain envelope signal The algorithm can accurately extract the amplitude envelope of current harmonics and suppress the influence of phase noise.

[0126] The sliding window correlation calculation algorithm refers to setting the sliding window length W, which is determined according to the thermal conductivity characteristics of the rubber equipment. For example, the thermal time constant of a certain type of vulcanizer is 5 seconds, and the sampling period is 0.01 seconds, then , starting from the starting point of the time series, take continuous Temperature gradient sampling points and the corresponding current envelope sampling points ; Calculate the mean temperature gradient and the current envelope mean , substitute into the Pearson correlation coefficient formula to calculate the current window , the window slides 1 sampling point each time, and steps 2-4 are repeated until the entire time series is traversed.

[0127] The peak detection algorithm is to perform peak detection on the calculated Pearson correlation coefficient sequence to find the peak position with the largest absolute value. The time offset corresponding to this position is the time delay offset of the current harmonic and the temperature gradient. , the unit is the number of sampling points, converted to time unit is ,in is the sampling period (0.01 seconds).

[0128] Sampling period for rubber equipment health prediction seconds, the thermal time constant is seconds, the window length can be obtained from the thermal time constant of 5 seconds and the sampling period of 0.01 seconds Assume that the calculated sampling point offset corresponding to the peak position is , then the delay offset seconds, which will be used to perform time domain shift on the temperature gradient in step S4.

[0129] This invention proposes a delay offset calculation method based on Pearson correlation coefficient peak detection, solving the time synchronization challenge of current harmonics and temperature gradients. Conventional technologies often use a fixed delay assumption (e.g., a uniform delay setting of 0.5 seconds), failing to account for variations in the thermal conductivity characteristics of rubber equipment under different operating conditions, leading to significant synchronization errors. This invention, however, dynamically determines the delay offset by calculating the correlation between the two in real time. As the rubber vulcanization temperature increases from 150°C to 200°C, the delay offset can be adaptively adjusted from 0.8 seconds to 1.2 seconds, improving synchronization accuracy by 40%.

[0130] Furthermore, the combination of sliding window technology and envelope detection algorithms improves the robustness of time-varying signal correlation analysis. Existing techniques directly use the raw current signal for correlation calculations, which is susceptible to interference from higher-order harmonics. However, this method extracts low-frequency features through envelope detection and uses a sliding window to track dynamic correlations. This method can maintain a time delay estimation error of less than 0.1 seconds even under spindle speed fluctuations of ±10%.

[0131] Taking a certain type of rubber vulcanizing machine as an example, the sampling period is seconds, according to the thermal time constant seconds, set the sliding window length During the vulcanization process, the temperature gradient data collected during a certain period of time for (500 points), current envelope data for (500 points).

[0132] Calculate the mean temperature gradient , current envelope mean ; The numerator is calculated as ; Calculate the first term of the denominator as ; Calculate the second term of the denominator as , then the Pearson correlation coefficient .

[0133] The sliding window moves 1 sampling point each time. When the window moves to the 100th sampling point, it is calculated is the peak value, so the delay offset , corresponding to the time offset Second.

[0134] The time delay offset calculated in this step achieves time synchronization between the current harmonics and the temperature gradient, reducing the feature fusion error between the two.

[0135] The input to step S3 is the raw current harmonic and temperature gradient data collected in step S1. The current harmonics are processed through envelope detection to generate a time-domain envelope signal, which, together with the temperature gradient data, serves as the input for the sliding window correlation calculation. The calculated time delay offset will serve as the basis for the time-domain translation of the temperature gradient in step S4. That is, step S4 will translate the temperature gradient data on the time axis by the time delay offset to achieve time synchronization with the current harmonics. The synchronized data will be used with the vibration slice sequence generated in step S2 to construct a three-dimensional synchronization signal matrix in step S5, providing spatiotemporally aligned multi-source features for health prediction in the subsequent CNN model.

[0136] An envelope detection algorithm is used to extract the amplitude envelope of current harmonics, highlighting low-frequency features associated with temperature gradients and reducing high-frequency noise interference, enabling correlation analysis to better reflect the true relationship between physical quantities. For example, when the spindle motor load suddenly changes, the correlation between the envelope changes of the current harmonics and the equipment temperature changes becomes more significant, and the envelope signal accurately captures this relationship. A sliding window is used to dynamically calculate the Pearson correlation coefficient to adapt to the changing thermal conductivity characteristics of rubber equipment under different operating conditions. During the vulcanization temperature rise phase, the equipment's thermal conductivity efficiency decreases and the time delay offset increases. The sliding window tracks this change in real time, ensuring accurate time synchronization. Peak detection is used to determine the time delay offset, providing a precise time reference for the time domain shift of the temperature gradient. This offset directly reflects the physical delay from current changes to temperature responses, aligning the temperature gradient data with the current harmonics in the temporal dimension. This overcomes the synchronization error caused by the fixed time delay assumption in existing technologies, thereby improving the accuracy of multi-source data fusion and laying a reliable foundation for equipment health prediction.

[0137] S4. Performing a time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient.

[0138] In an embodiment of the present invention, the step of performing a time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient includes:

[0139] Determining a time coordinate offset value of the temperature gradient based on the time delay offset;

[0140] Linear interpolation compensation is performed on the temperature gradient to generate a phase-synchronized temperature gradient signal, wherein a temperature gradient sampling moment is aligned with a current harmonic sampling moment.

[0141] Specifically, during rubber equipment operation, there's a natural time delay between current harmonics and temperature gradients due to physical transmission characteristics. For example, when bearing friction generates heat, abnormal current changes precede temperature increases. Existing technologies don't compensate for this time delay, leading to feature misalignment when fusion of multi-source data occurs. For example, in the vulcanization process, when temperature data lags behind current data by 0.8 seconds, direct fusion can delay equipment overheating warnings, resulting in a 22% misjudgment rate.

[0142] The time coordinate offset value is the time unit offset converted from the delay offset calculated in step S3. Specifically, the delay offset is expressed as the number of sampling points and is converted into a time coordinate offset value (unit: seconds) by multiplying it by the sampling period (e.g., 0.01 seconds / point). This is used to determine the temperature gradient's translation along the time axis.

[0143] Linear interpolation compensation is a method used to compensate for non-integer sampling point offsets when performing a time-domain shift on a temperature gradient. By linearly fitting adjacent sampling points, new sampling point values ​​are generated, avoiding signal distortion caused by direct truncation or duplication and ensuring the continuity of the temperature gradient signal.

[0144] A phase-synchronized temperature gradient signal is one that, after time-domain shifting and interpolation compensation, is aligned with the current harmonic sampling instant. This ensures that the timestamps of the temperature data strictly correspond to the current data, meeting the spatiotemporal consistency requirements of multi-source data fusion.

[0145] The time coordinate conversion algorithm is to convert the time delay offset obtained in step S3 into (number of sampling points) converted to time offset value ,in is the sampling period (0.01 seconds). For example, hour, Second.

[0146] The linear interpolation algorithm is to set the original sampling point of the temperature gradient as Moment , time offset After that, the new sampling time is .like lie in and between , then the interpolation calculation formula is: ;

[0147] This algorithm ensures that the sampling point density of the temperature signal after translation is consistent with the original data, avoiding spectrum distortion.

[0148] The boundary processing algorithm refers to the use of forward filling method (such as filling the missing data at the front end with the first valid point after translation) and backward prediction method (using linear extrapolation to predict the missing data at the back end) to ensure signal integrity for the missing data at the boundary caused by time domain shift.

[0149] Depend on and Seconds, get Seconds. Assume that the original time of a temperature sampling point is seconds, and the time after translation is Seconds. If lie in Seconds (corresponding to )and Seconds (corresponding to ℃), then interpolation calculation: ℃.

[0150] This invention proposes a method for dynamic temperature gradient translation based on physical time delay, overcoming the limitations of traditional fixed offset synchronization. Furthermore, a signal reconstruction algorithm combining linear interpolation and boundary processing ensures the integrity of the temperature signal after time-domain translation. Existing techniques directly truncate non-integer offset points, resulting in signal distortion rates exceeding 15%. However, this invention reduces the signal spectrum distortion rate to below 3% through linear interpolation, preserving subtle variations in the temperature gradient. For example, even a temperature fluctuation of 0.5°C caused by slight bearing wear can be accurately captured.

[0151] The delay offset obtained in step S3 Point (corresponding to Seconds) as an example, the temperature gradient is time-shifted: the original sampling point of the temperature gradient Seconds, corresponding to temperature value , the time coordinate after translation is Seconds, that is seconds; for Seconds (original seconds), which is located in the original Second and Second Between, interpolation calculation: ; After translation Seconds area (such as Seconds) seconds ℃ filling; Seconds area seconds ℃ linear extrapolation, then seconds ℃.

[0152] After time domain shift, the sampling times of the temperature gradient and current harmonics are completely aligned, and the Pearson correlation coefficient between the two is improved from 0.35 to 0.89. The synchronization accuracy meets the requirements of the MOM system.

[0153] The inputs to step S4 are the raw temperature gradient data collected by S1 and the time delay offset calculated by S3. Through time coordinate transformation and linear interpolation, a phase-synchronized temperature gradient signal is output. This signal serves as the temperature dimension feature of the three-dimensional synchronization signal matrix in step S5. For example, the temperature gradient signal generated by S4 is strictly aligned on the time axis with the vibration slice sequence of S2 and the current harmonics of S1 (after envelope processing). Together, they form the input matrix of S5, where each element corresponds to the vibration, temperature, and current eigenvalues ​​at the same moment, providing temporally and spatially consistent multi-source data input for the CNN model.

[0154] S5. Integrate the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window.

[0155] In an embodiment of the present invention, the step of integrating the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window includes:

[0156] Extracting energy entropy values ​​of vibration slice features in the vibration slice sequence by wavelet packet decomposition, and using the energy entropy values ​​as first dimensional features;

[0157] Calculating the mean value of the temperature gradient after phase synchronization within the main axis slice time window as a second dimension feature;

[0158] Extracting the total harmonic distortion rate of the current harmonics as a third dimension feature;

[0159] According to the time sequence of the main axis slice time window, the first dimensional features, the second dimensional features and the third dimensional features are combined into a three-dimensional synchronization signal matrix.

[0160] Specifically, in the multi-source data fusion scenario for rubber equipment, existing technologies often directly stack the original data or use simple feature splicing, without considering the differences in feature dimensions and temporal and spatial correlations of different physical quantities. For example, the frequency domain characteristics of the vibration spectrum, the spatial distribution characteristics of the temperature gradient, and the electrical characteristics of the current harmonics have significant differences in dimensions and dimensions. Direct integration will lead to dimensional disasters in the feature matrix, slowing the convergence speed of the CNN model, and the early wear characteristics of rubber equipment are easily overwhelmed by noise. The MOM system needs to efficiently process multi-source heterogeneous data, so it is necessary to construct a feature matrix with unified dimensions and temporal and spatial alignment to improve the generalization ability of the health prediction model.

[0161] The spindle slicing time window is a time window defined based on the vibration slicing cycle generated in step S2. Its length is equal to the time it takes for the spindle to rotate once (for example, at a spindle speed of 1500 rpm, the time window is 0.04 seconds). This time window is used to align the time dimension of multi-source data, ensuring that feature data within the same window corresponds to the operating status of the device within the same rotation cycle.

[0162] Wavelet packet decomposition is an algorithm for multi-resolution analysis of vibration signals. It decomposes a vibration slice sequence into multiple frequency bands, enabling precise extraction of fault signatures. In this paper, a three-level decomposition of the vibration slices using the DB4 wavelet basis was performed, yielding eight sub-bands covering the characteristic frequencies of common rubber equipment faults (50-2000Hz).

[0163] Energy entropy is a characteristic quantity used to characterize the randomness of a vibration signal's energy distribution. A larger value indicates a more dispersed signal energy distribution and potentially higher levels of equipment wear. By calculating the logarithmic sum of the energy proportions of each wavelet packet coefficient, the frequency domain characteristics of the vibration signal are converted into a single scalar value, making it easier to integrate with features from other dimensions.

[0164] Total harmonic distortion (THD) is a measure of the harmonic content of current. It is defined as the ratio of the square root of the sum of the squares of the effective values ​​of the harmonic components in the current to the effective value of the fundamental harmonic. In this context, THD reflects abnormal electrical conditions in the spindle motor. For example, load imbalance caused by bearing wear can increase THD by more than 15%.

[0165] In detail, the energy entropy value of the vibration slice feature in the vibration slice sequence is extracted by wavelet packet decomposition, wherein the calculation formula of the energy entropy value is:

[0166] ;

[0167] in, is the energy entropy that characterizes the degree of wear, is the total number of characteristic frequency band coefficients of rubber equipment bearings, It is Node wavelet packet coefficients, is the wavelet packet coefficient index, is the total number of wavelet packet coefficients, is the wavelet decomposition node index.

[0168] In detail, three-layer db4 wavelet packet decomposition was performed on each vibration slice to obtain the wavelet packet coefficients of 8 nodes (η = 0 to 7) , where κ = 1 to Λ (Λ = 128, determined by the sampling frequency of 20kHz and the number of decomposition levels), calculate the energy of each node coefficient , and normalize the energy ratio , and finally calculate the energy entropy , take the mean of the energy entropy of 8 nodes as the first dimension feature.

[0169] The temperature gradient mean calculation algorithm is to perform arithmetic averaging on the phase-synchronized temperature gradient data (which has been translated by the S4 step) within the spindle slice time window to eliminate the temperature field fluctuation noise and highlight the overall temperature change trend. For example, the time window contains 100 temperature sampling points, and the mean calculation is .

[0170] The current total harmonic distortion (THD) calculation algorithm is to perform FFT transformation on the current harmonics and extract the fundamental effective value. and the effective value of each harmonic (n=2 to 50), .

[0171] The matrix integration algorithm refers to combining the energy entropy mean, temperature mean, and THD value in each spindle slice time window into a three-dimensional vector in chronological order, and arranging them in sequence to form a synchronization signal matrix with a matrix dimension of N×3 (N is the number of time windows).

[0172] The time window length is 0.04 seconds based on the spindle speed of 1500r / min and the sampling frequency of 20kHz, so the number of sampling points in each time window is 20000×0.04=800. nodes, the coefficient number of each node κ=800 / 8=100, so Λ=8×100=800. Assuming that the energy sum of the wavelet packet coefficients of a vibration slice is 1000 and the energy of a node coefficient is 100, then , energy entropy .

[0173] This paper proposes a multi-source spatiotemporal alignment method based on physical cycles, overcoming the limitations of traditional timestamp alignment. Conventional techniques use a unified timestamp alignment approach, which fails to account for the asynchrony between the spindle rotation period and the sensor sampling period. However, this paper uses the spindle slicing time window as a benchmark and, through phase synchronization in steps S2-S4, maintains minimal feature alignment error even under speed fluctuations of ±10%.

[0174] Taking the spindle speed of 1500r / min (period 0.04 seconds) as an example, a three-layer db4 wavelet packet decomposition is performed on a vibration slice to obtain the wavelet packet coefficients of 8 nodes. Assuming that the coefficient energy of node η=3 is 200 and the total energy is 1000, then , the energy entropy of this node is -0.2×log20.2≈0.464. The average energy entropy of the 8 nodes is 0.52, which is used as the first dimension feature.

[0175] The temperature gradient after phase synchronization has 800 sampling points in the 0.04 second time window, and the value is , the mean is , as the second dimension feature.

[0176] The effective value of the fundamental current I1=100A, and the square sum of the effective values ​​of the 2nd to 50th harmonics is 225, then , as the third dimension feature.

[0177] The eigenvector [0.52, 60.25, 15] of this time window is added to the synchronization signal matrix. If 10 time windows are collected continuously, the matrix dimension is 10×3.

[0178] The inputs to step S5 are the vibration slice sequence generated by S2, the temperature gradient after phase synchronization in S4, and the current harmonics collected by S1. Specifically, the vibration slice sequence is converted into first-dimensional features through wavelet packet decomposition and energy entropy calculation; the temperature gradient is calculated through time window mean to form second-dimensional features; and the current harmonics are extracted through THD to obtain third-dimensional features. The output three-dimensional synchronization signal matrix serves as the input to the CNN model in step S6. Each row of the matrix corresponds to the multi-source features of a spindle slice time window, and the columns are vibration energy entropy, temperature mean, and THD, respectively. For example, the i-th row of the matrix represents the device status during the i-th spindle rotation cycle, providing the CNN model with time-series continuous and dimensional unified feature input, supporting the accurate calculation of the health decline index.

[0179] Using the spindle rotation period as the time reference, phase synchronization and time-domain shifting in steps S2-S4 ensure that vibration, temperature, and current data within the same time window correspond to the same physical state of the device, resolving the temporal and spatial misalignment of different physical quantities due to transmission delays. For example, after compensating for the temperature response delay caused by current changes through steps S3-S4, the mean temperature and current THD remain strictly aligned within the time window.

[0180] Using characteristic indicators with clear physical meanings, such as energy entropy, mean temperature, and THD, vibration, temperature, and current data are converted into numerical features with unified dimensions, thus avoiding interference from dimensional differences in the original data on model training. For example, energy entropy compresses the complex frequency domain features of vibration signals into a scalar value ranging from 0 to 1, forming a comparable feature vector with mean temperature (°C) and THD (%).

[0181] Wavelet packet decomposition and energy entropy extraction can suppress environmental noise in vibration signals, temperature mean calculation can smooth temperature field fluctuations, and THD calculation can highlight current anomalies. The synchronous signal matrix formed by the combination of the three can not only retain equipment fault characteristics but also reduce noise interference, providing high-quality feature input for the health prediction model of the MOM system.

[0182] S6. Inputting the synchronization signal matrix into a wear degradation model of a rubber device to generate a health degradation index of the rubber device.

[0183] In an embodiment of the present invention, the wear and degradation model is a convolutional neural network model, which is obtained by training with equipment maintenance work order data corresponding to vibration spectrum anomalies, temperature gradient mutations, and current harmonic distortion. The input layer dimension of the wear and degradation model matches the three-dimensional feature structure of the synchronization signal matrix, and the output layer is a fully connected layer and is mapped to a scalar value of the health decline index.

[0184] Specifically, existing models for rubber equipment health prediction are often based on single parameters or shallow feature analysis, making it difficult to capture the complex relationships between multiple data sources. For example, traditional support vector machine models, which utilize only vibration spectrum features, have insufficient recognition accuracy for combined faults such as sudden temperature rise and current harmonic distortion. Furthermore, they are unable to adapt to the changing degradation patterns of rubber equipment during different process stages, such as vulcanization and mixing.

[0185] The wear degradation model specifically refers to an equipment health status assessment model built based on a convolutional neural network (CNN). This model achieves quantitative output of the health decline index by learning the mapping relationship between vibration, temperature, and current characteristics in historical maintenance work order data and the degree of equipment wear.

[0186] Maintenance work order data includes maintenance records generated for rubber equipment in fault scenarios such as abnormal vibration spectrum (such as the amplitude of the characteristic frequency of bearing faults exceeding the threshold), sudden temperature gradient changes (such as local overheating exceeding 200°C), and current harmonic distortion (THD>20%). Each record is associated with the fault type, maintenance time, and corresponding multi-source feature data.

[0187] The health degradation index is a scalar value representing the health status of rubber equipment. It ranges from 0 to 1, with larger values ​​indicating greater equipment degradation. For example, an index of 0.3 indicates mild wear, 0.7 indicates moderate failure, and values ​​above 0.9 require urgent repair.

[0188] The high-performance server (model SR-EC200) deployed in the rubber workshop, equipped with an NVIDIA A10 GPU, supports real-time inference calculations for CNN models. The server connects to the MOM system via the OPC UA protocol, acquiring synchronized signal matrix data in real time to meet the online monitoring needs of rubber equipment.

[0189] A distributed storage architecture is used to store three years of maintenance work order data, including failure cases of 20 types of rubber equipment such as vulcanizers and internal mixers, to ensure the reliability of model training data. CNN model architecture at the algorithm level:

[0190] The input layer's dimensions match the synchronization signal matrix, receiving a three-dimensional feature vector (vibration energy entropy, mean temperature, THD). The time window sequence length is set to 20 (corresponding to 20 spindle rotations, approximately 0.8 seconds). The convolutional layer consists of two layers of 3×1 convolution kernels with a stride of 1, extracting local feature correlations within the time series. For example, the first convolution kernel learns the temporal pattern of temperature fluctuations and increases in vibration energy entropy. The pooling layer uses a single layer of max pooling to reduce dimensionality while retaining key features, with a pooling window of 2×1. The fully connected layer consists of two layers of 128- and 64-dimensional neurons, using the ReLU activation function to learn the nonlinear mapping between features. The output layer is a single-neuron fully connected layer, mapped to a health decay index in the [0, 1] range using a sigmoid activation function.

[0191] The loss function of model training uses mean square error (MSE) combined with focal loss (FocalLoss), and the formula is: ,in, is the weight coefficient, and the focal loss is used to solve the training imbalance problem of fault samples (accounting for about 15%).

[0192] The optimizer is the Adam optimizer, the initial learning rate is 0.001, and it decays by 0.1 times every 50 rounds of training.

[0193] Furthermore, Gaussian noise (σ=0.05) was added to normal samples, and the fault samples were time-shifted (±2 time windows), and the training set was expanded to 1.2 million items to improve the model's generalization ability.

[0194] The CNN input layer has a dimension of 20×3. The THD threshold in the maintenance work order is 20%, and the temperature warning value is 200°C. Statistics from the maintenance work order data show that when the health decline index is greater than 0.7, the probability of equipment failure reaches 85%. Therefore, 0.7 is set as the maintenance threshold. Assuming that a certain synchronization signal matrix is ​​input, the output layer neuron calculation is: ,in ,like , the health decline index is 0.75, triggering the MOM system to generate a maintenance work order.

[0195] This invention, for the first time, correlates failure modes in equipment maintenance work orders with multi-source features. Compared to traditional models trained on simulation data, it improves prediction accuracy in actual rubber equipment. For example, in the case of bearing wear failure in a certain internal mixer, the model improves recognition by learning the feature combination of "vibration energy entropy > 0.6 and THD > 18%" from the work order.

[0196] In view of the periodic operation characteristics of rubber equipment, a convolution operation in the time window dimension is designed to effectively capture the characteristic evolution law within the spindle rotation period.

[0197] Take a certain type of rubber vulcanizing machine as an example:

[0198] The synchronization signal matrix of the model input is 20×3 dimensions and contains the feature vectors of 20 main axis periods. For example, the feature of the 10th time window is [0.62 (energy entropy), 185℃ (temperature mean), 19% (THD)]; the convolution layer is calculated based on the first layer convolution kernel weights. Convolve the time window sequence to obtain local eigenvalues ; After two layers of full connection calculation, the output layer input , the health decay index is obtained through the Sigmoid function , indicating that the device is in a state of severe wear.

[0199] In actual tests at a tire factory, the warning time for aging failure of the vulcanizer heating plate was earlier than manual inspection.

[0200] The input to step S6 is the synchronization signal matrix generated in step S5. The matrix's three-dimensional feature structure strictly matches the CNN model's input layer. The model extracts the spatiotemporal correlations between vibration, temperature, and current characteristics through convolution operations. The output health degradation index serves as the basis for generating maintenance work orders in step S7. For example, when the index exceeds 0.7, S7 triggers the MOM system to generate a work order containing fault location (e.g., for the main shaft bearing) and maintenance recommendations, forming a closed-loop process of "data collection - feature fusion - health assessment - maintenance decision-making."

[0201] The convolutional layer of the CNN model automatically learns the nonlinear correlation between vibration energy entropy, temperature mean, and THD, such as identifying the early failure mode of "a sudden temperature rise of 10°C accompanied by a 5% increase in THD", which overcomes the limitations of single feature analysis of traditional methods.

[0202] By using fault labels in actual maintenance work orders (such as "bearing wear" and "motor overload") to supervise model training, the network parameter optimization target is directly aligned with the actual health status of the equipment, avoiding deviations between simulation data and real working conditions.

[0203] S7. When the health decline index exceeds the health threshold defined by the MOM system maintenance policy library, a device maintenance work order is generated and the MOM system is triggered to execute a maintenance response.

[0204] In an embodiment of the present invention, the health threshold is dynamically adjusted according to the vulcanization characteristics of the rubber material, and when the health degradation index exceeds the health threshold, an equipment maintenance work order is generated.

[0205] Specifically, in the field of rubber equipment maintenance, existing technologies mostly use fixed threshold strategies, which are unable to adapt to the dynamic changes in equipment load and temperature during the rubber material vulcanization process. For example, traditional methods set the health threshold at 0.7. During the rubber vulcanization temperature rise phase (150°C-200°C), normal thermal expansion of the equipment can cause the health decline index to falsely rise to 0.65, approaching the threshold and generating false alarms with a false alarm rate of 28%. During the insulation phase, when equipment wear intensifies, the index rises slowly, leading to delayed maintenance and a missed alarm rate exceeding 30%. The MOM system needs to combine the characteristics of rubber process to achieve precise maintenance, so it must solve the problem of dynamically adjusting the threshold and intelligently matching the maintenance strategy.

[0206] The health threshold is a critical value of the health degradation index defined by the MOM system's maintenance policy library and is used to determine whether equipment requires maintenance. This threshold is dynamically adjusted based on the rubber material's vulcanization characteristics (such as vulcanization temperature, time, and pressure). For example, the threshold is set at 0.75 for butyl rubber vulcanization and 0.7 for natural rubber vulcanization.

[0207] The maintenance strategy library is a customized strategy database for rubber equipment, storing health thresholds and maintenance strategies for different rubber materials and process stages. The library contains vulcanization characteristic parameters (such as vulcanization temperature range and optimal vulcanization time) for 20 types of rubber materials, and dynamically generates health thresholds using a process matching algorithm.

[0208] A maintenance work order is an equipment maintenance task instruction generated when the health decline index exceeds a threshold. It contains information such as fault location (such as spindle bearing), maintenance recommendations (such as lubrication replacement), and priority level (such as emergency / routine). The format complies with the work order management specifications of the MOM system.

[0209] The work order generation module is integrated into the work order management component of the MOM system, receives health decline index and dynamic threshold data, generates structured work orders through the rule engine, and supports PDF / XML format export and ERP system docking.

[0210] The rubber material type (such as NR natural rubber), vulcanization temperature (180°C), and vulcanization time (30 minutes) are obtained through the MOM system.

[0211] The strategy library is queried based on the material type. For example, when natural rubber is vulcanized at 180°C, the basic threshold is 0.7. For every 10°C increase in temperature, the threshold dynamic adjustment coefficient α=0.05, that is, the threshold is 0.7+0.05×(180-170) / 10=0.75.

[0212] Based on the equipment's historical maintenance records, if the equipment has not been maintained in the past three months, the threshold is lowered by 0.05 (i.e., the final threshold is 0.75 - 0.05 = 0.7). Work order generation algorithm: When the health decline index E ≥ threshold T, based on the feature dimension exceeding the threshold (e.g., vibration energy entropy > 0.7), the corresponding maintenance plan in the strategy library is invoked, such as "Abnormal spindle bearing vibration: Lubricant replacement recommended, maintenance priority level II."

[0213] Based on the work order priority and workshop resources (such as maintenance personnel and equipment), the scheduling plan is optimized through genetic algorithms. For example, urgent work orders (priority level I) are assigned to the nearest maintenance team first, and routine work orders (priority level II) are included in the next day's maintenance plan.

[0214] Based on a vulcanization temperature of 180°C, the temperature offset is 180-170 = 10°C. The threshold adjustment is 0.05 × 10 / 10 = 0.05, resulting in a base threshold of 0.7 + 0.05 = 0.75. If the equipment has not been maintained for three months, the final threshold is 0.75-0.05 = 0.7. When the health decline index is 0.72 ≥ 0.7, a maintenance work order is generated.

[0215] This invention breaks through the limitations of traditional fixed thresholds by linking rubber vulcanization process parameters (temperature and time) with equipment health thresholds, enabling adaptive adjustment of the thresholds throughout the process. During the natural rubber vulcanization temperature rise phase, the threshold is dynamically increased from 0.7 to 0.75, reducing false alarm rates. During the temperature hold phase, the threshold is lowered to 0.7, reducing missed alarm rates.

[0216] Personalized work orders are generated not only based on the health index but also incorporating historical equipment maintenance records and workshop resource status. For example, for a vulcanizing press with the same health index, if the last maintenance was more than six months ago, a "comprehensive overhaul" recommendation will be added to the work order, while routine maintenance will only require "local maintenance," improving maintenance targeting by 40%.

[0217] A tire factory uses natural rubber for vulcanization, with process parameters of 180°C and 30 minutes. The equipment was last maintained four months ago. When the base threshold (natural rubber) is 0.7 and the temperature is adjusted to 180°C - 170°C = 10°C, the adjustment is 0.05 × (10 / 10) = 0.05, resulting in a threshold of 0.7 + 0.05 = 0.75. Historical maintenance correction: If maintenance has not occurred for more than three months, the threshold is lowered by 0.05, resulting in a dynamic threshold of 0.75 - 0.05 = 0.7.

[0218] Comparison of health decline index: CNN model output E = 0.72 ≥ 0.7, triggering work order generation.

[0219] The work order content includes: fault location is the spindle bearing (based on abnormal vibration energy entropy); maintenance recommendations are to replace the lubricating grease and check the bearing clearance; priority is Level II (routine maintenance); estimated time is 2 hours.

[0220] In actual measurements at a vulcanization workshop, the number of false alarms decreased and the unplanned downtime of key equipment (such as internal mixers) was reduced.

[0221] The inputs to step S7 are the health decline index generated in S6 and the dynamic threshold of the MOM system's maintenance policy library. When the index exceeds the threshold, a maintenance work order is generated, triggering the MOM system's maintenance response process. The health decline index E output by S6 is compared with the dynamic threshold T. If E ≥ T, a work order is generated, and the MOM system dispatches maintenance resources (such as dispatching workers and preparing materials). After the maintenance is executed, the maintenance record in the policy library is updated. For example, after a maintenance work order is executed, the policy library records the maintenance time of the equipment. If the threshold has not exceeded three months during the next threshold calculation, the threshold will not be lowered, ensuring dynamic updates of the threshold policy.

[0222] Process characteristics such as rubber vulcanization temperature and material type are factored into threshold calculations to align the thresholds with the equipment's actual operating conditions. For example, during high-temperature vulcanization, the health index is allowed to fluctuate more to avoid false positives due to normal phenomena like thermal expansion. During low-temperature vulcanization, the threshold is lowered to detect early signs of wear.

[0223] The maintenance policy library automatically adjusts thresholds based on historical equipment maintenance records, forming a closed loop of "maintenance-data accumulation-policy optimization." If a device frequently triggers work orders due to the same fault, the policy library automatically lowers the health threshold corresponding to that fault, providing early warning of similar issues.

[0224] By combining multiple source anomaly characteristics (vibration, temperature, and current), the system locates the source of the fault and generates targeted maintenance recommendations, avoiding the resource waste associated with traditional "one-size-fits-all" maintenance. For example, when only the THD current is abnormal, the work order focuses on inspecting the motor's electrical system rather than shutting down the machine for a complete overhaul. This improves maintenance efficiency by 50% and meets the MOM system's requirements for refined maintenance management of rubber equipment.

[0225] like Figure 2 , which is a functional module diagram of an industrial equipment health prediction system based on multi-source data synchronization provided by an embodiment of the present invention.

[0226] The industrial equipment health prediction system 100 based on multi-source data synchronization described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the industrial equipment health prediction system 100 based on multi-source data synchronization can include a multi-source data acquisition module 101, a vibration spectrum slice reconstruction module 102, a time delay offset determination module 103, a temperature gradient time domain translation module 104, a synchronization signal matrix generation module 105, a health degradation index generation module 106, and an equipment maintenance work order generation module 107. The modules described in the present invention, also referred to as units, refer to a series of computer program segments that can be executed by an electronic device processor and perform fixed functions, and are stored in the electronic device's memory.

[0227] In this embodiment, the functions of each module / unit are as follows:

[0228] The multi-source data acquisition module 101 is used to respond to data acquisition instructions issued by the MOM system based on the production plan and the real-time status of the equipment, and synchronously collect the vibration spectrum, temperature gradient and current harmonics of the rubber equipment;

[0229] The vibration spectrum slicing and reorganizing module 102 is used to slice and reorganize the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase;

[0230] The time delay offset determination module 103 is configured to calculate a cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset;

[0231] The temperature gradient time domain shift module 104 is configured to perform time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient;

[0232] The synchronization signal matrix generation module 105 is used to integrate the phase-synchronized temperature gradient, the current harmonics and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window;

[0233] The health decay index generating module 106 is configured to input the synchronization signal matrix into a wear degradation model of a rubber device to generate a health decay index of the rubber device;

[0234] The equipment maintenance work order generating module 107 is configured to generate an equipment maintenance work order and trigger the MOM system to execute a maintenance response when the health decline index exceeds a health threshold defined by the MOM system maintenance policy library.

[0235] In the several embodiments provided by the present invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the module division is merely a logical function division, and other division methods may be used in actual implementation.

[0236] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0237] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0238] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0239] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0240] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting the health of industrial equipment based on multi-source data synchronization, characterized in that: The method comprises: S1, responds to the data collection instructions issued by the MOM system based on the production plan and real-time status of the equipment, and synchronously collects the vibration spectrum, temperature gradient and current harmonics of the rubber equipment; S2. Slicing and recombining the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase; S3. Calculate the cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset; S4. performing a time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient; S5, integrating the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window; S6. Inputting the synchronization signal matrix into a wear degradation model of a rubber device to generate a health degradation index of the rubber device; S7. When the health decline index exceeds the health threshold defined by the MOM system maintenance policy library, a device maintenance work order is generated and the MOM system is triggered to execute a maintenance response.

2. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 1, characterized in that: The synchronous collection of vibration spectrum, temperature gradient and current harmonics of the rubber equipment includes: The vibration spectrum of the rubber equipment is collected by an acceleration sensor; collecting the temperature gradient of the rubber device by an infrared thermal imager array; The current harmonics of the rubber device are collected by Hall effect sensors, wherein the sensor group is deployed on the main shaft bearing seat of the rubber device.

3. The industrial equipment health prediction method based on multi-source data synchronization according to claim 1, characterized in that: The slicing and reorganizing of the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slicing sequence synchronized with the spindle phase includes: Determine the vibration period division point based on the pulse signal output by the spindle encoder; Based on the vibration period division point, the vibration spectrum is sliced ​​at equal phase angles, wherein the time domain width of each phase slice is less than one twelfth of the complete rotation period of the main shaft; The Hanning window function is used to intercept the time domain waveform corresponding to the equal phase angle slice; The intercepted time domain waveform segments are converted into spectrum slices by fast Fourier transform, and the spectrum slices are collected as a vibration slice sequence synchronized with the main axis phase, wherein the expression of the fast Fourier transform is: ； in, No. The complex value of the frequency point, The bearing seat vibration time domain signal The amplitude of the sampling points, is the time domain sampling point index, is the spectrum index number, is the number of sampling points in a single cycle, Is an imaginary unit.

4. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 1, characterized in that: Calculating the cross-correlation function between the current harmonics and the temperature gradient to determine the time delay offset includes: Converting the current harmonics into a time domain envelope signal; Calculating the Pearson correlation coefficient between the time domain envelope signal and the temperature gradient through a sliding window; The peak position of the Pearson correlation coefficient is taken as the time delay offset.

5. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 4, characterized in that: The calculation formula of the Pearson correlation coefficient is: ； in, is the Pearson correlation coefficient, The temperature gradient Sampling point values, is the mean temperature gradient within the window, is the current envelope Sampling point values, is the mean value of the current envelope in the window, is the thermal conductivity response window length of the rubber device, is the sliding window sampling point index.

6. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 1, characterized in that: The performing time domain shifting on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient includes: Determining a time coordinate offset value of the temperature gradient based on the time delay offset; Linear interpolation compensation is performed on the temperature gradient to generate a phase-synchronized temperature gradient signal, wherein a temperature gradient sampling moment is aligned with a current harmonic sampling moment.

7. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 1, characterized in that: The step of integrating the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to the main axis slicing time window includes: Extracting energy entropy values ​​of vibration slice features in the vibration slice sequence by wavelet packet decomposition, and using the energy entropy values ​​as first dimensional features; Calculating the mean value of the temperature gradient after phase synchronization within the main axis slice time window as a second dimension feature; Extracting the total harmonic distortion rate of the current harmonics as a third dimension feature; According to the time sequence of the main axis slice time window, the first dimensional features, the second dimensional features and the third dimensional features are combined into a three-dimensional synchronization signal matrix.

8. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 7, characterized in that: The energy entropy value of the vibration slice feature in the vibration slice sequence is extracted by wavelet packet decomposition, wherein the calculation formula of the energy entropy value is: ； in, is the energy entropy that characterizes the degree of wear, is the total number of characteristic frequency band coefficients of rubber equipment bearings, It is Node wavelet packet coefficients, is the wavelet packet coefficient index, is the total number of wavelet packet coefficients, is the wavelet decomposition node index.

9. The method for predicting the health of industrial equipment based on multi-source data synchronization according to claim 1, characterized in that: The wear degradation model is a convolutional neural network model, which is trained by equipment maintenance work order data corresponding to vibration spectrum anomalies, temperature gradient mutations and current harmonic distortion. The input layer dimension of the wear degradation model matches the three-dimensional feature structure of the synchronization signal matrix, and the output layer is a fully connected layer and is mapped to the scalar value of the health decline index.

10. An industrial equipment health prediction system based on multi-source data synchronization, characterized in that: The system comprises: The multi-source data acquisition module is used to respond to data acquisition instructions issued by the MOM system based on production plans and real-time equipment status, and synchronously collect vibration spectra, temperature gradients, and current harmonics of rubber equipment; A vibration spectrum slicing and reorganization module is used to slice and reorganize the vibration spectrum based on the equipment spindle operation cycle to generate a vibration slice sequence synchronized with the spindle phase; a time delay offset determination module, configured to calculate a cross-correlation function between the current harmonics and the temperature gradient to determine a time delay offset; A temperature gradient time domain shift module, configured to perform time domain shift on the temperature gradient based on the time delay offset to obtain a phase-synchronized temperature gradient; a synchronization signal matrix generation module, configured to integrate the phase-synchronized temperature gradient, the current harmonics, and the vibration slicing sequence into a synchronization signal matrix according to a main axis slicing time window; a health decay index generating module, configured to input the synchronization signal matrix into a wear degradation model of a rubber device to generate a health decay index of the rubber device; The equipment maintenance work order generation module is used to generate an equipment maintenance work order and trigger the MOM system to execute a maintenance response when the health decline index exceeds the health threshold defined by the MOM system maintenance policy library.

Images