A method and system for managing the entire lifecycle information of power plant equipment

Abstract

This application provides a method and system for managing the entire lifecycle information of power plant equipment, relating to the field of equipment condition monitoring technology. The method includes: acquiring oil index data and vibration data associated with QR code tags; mapping the elemental content in the oil to the source of wear components based on the material composition of equipment parts to construct a wear distribution map; identifying the dominant wear components based on the wear distribution map and extracting state feature vectors; combining the state feature vectors with vibration data and using a particle filter algorithm for state tracking, recursively predicting the remaining lifespan distribution; comparing the remaining lifespan distribution with the planned maintenance window, and pushing maintenance suggestions and writing them into the database when the remaining lifespan is lower than the planned maintenance interval. This application improves the synergy between wear condition assessment and maintenance decision-making for rotating equipment.

Description

Technical Field

[0001] This application relates to the field of equipment condition monitoring technology, and in particular to a method and system for managing information throughout the entire life cycle of power plant equipment. Background Technology

[0002] In the power industry, condition monitoring and full life-cycle management of power plant equipment are core components for ensuring power generation safety and reducing operation and maintenance costs. As equipment service life increases, accurate prediction of remaining life and maintenance decisions for rotating machinery have significant engineering application value.

[0003] In existing equipment management technologies, two independent methods are typically used for fault diagnosis: oil analysis and vibration detection. Oil analysis is a method that indirectly determines the internal wear condition of equipment by detecting the content of wear-prone metal elements in the lubricating oil. It reflects the cumulative chemical information of wear within the equipment, but the sampling period is long, making it difficult to capture sudden faults in real time. Vibration detection is a method that identifies fault modes by analyzing the mechanical vibration signals generated during equipment operation. It reflects the physical information of the equipment's current operating state and can continuously monitor but is difficult to identify the specific component source of wear.

[0004] Traditional management models often process these two types of data separately, only performing correlation analysis when obvious anomalies occur. Furthermore, maintenance plans are primarily based on equipment operating time or fixed time intervals, lacking the ability to adaptively adjust to the real-time health status of the equipment. Because oil level monitoring data and vibration signals inherently differ in time scale, direct splicing or simple superposition fails to accurately capture the dynamic evolution of equipment status, leading to significant discrepancies between remaining life predictions and the actual health condition of the equipment. Maintenance decisions are either overly conservative, resulting in resource waste, or overly aggressive, causing operational risks. Therefore, existing technologies suffer from insufficient coordination between equipment condition prediction and maintenance plans. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the present invention aims to provide a method and system for managing the entire lifecycle information of power plant equipment, thereby solving the problem of poor coordination between wear condition assessment and maintenance decision-making for rotating equipment in the prior art.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for managing the entire lifecycle information of power plant equipment, comprising: The oil index data and vibration data associated with the QR code label at the sampling port of the device are obtained. The oil index data includes the content of multiple elements in the lubricating oil sample, and the acquisition time of the vibration data is synchronized with the sampling time of the lubricating oil sample. Based on the material composition of different components in the rotating equipment of the power plant, the different contents of the aforementioned elements are mapped to the corresponding wear component sources to construct a wear distribution map, which is used to characterize the wear degree of each wear component inside the equipment; The wear-dominant component of the equipment is determined based on the wear distribution map, and the state feature vector corresponding to the wear-dominant component is extracted from the wear distribution map. The state feature vector is combined with the vibration data. Based on the correspondence between the sampling time and the acquisition time, the particle filter algorithm is used to perform state tracking on the combined data to recursively predict the wear state evolution trend of the equipment in future operating cycles and output the remaining life distribution. The remaining lifetime distribution is compared with the planned maintenance window in the lifecycle management database. When the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, a maintenance suggestion is pushed through the mobile terminal and written into the lifecycle management database associated with the QR code label.

[0007] Optionally, the step of combining the state feature vector with the vibration data, and using a particle filter algorithm to perform state tracking on the combined data based on the correspondence between the sampling time and the acquisition time, to recursively predict the wear state evolution trend of the equipment in future operating cycles and output the remaining lifetime distribution, includes: Each fitting coefficient in the state feature vector is mapped to a different phase angle in the complex plane to generate a set of complex feature components; Each vibration amplitude in the vibration data is taken as the real part and multiplied by the complex feature component at the corresponding sampling time point to obtain a complex joint vector; Using the correspondence between the sampling time and the acquisition time as the time axis, the complex joint vectors at each time point are arranged in ascending order of time to construct a complex state sequence; The real and imaginary parts of the complex state sequence are used as two parallel input channels and input into the particle filtering algorithm. The particle filtering algorithm outputs the real and imaginary trajectories of each particle. The modulus of the real and imaginary trajectories of each particle is calculated to obtain the remaining running time for each particle. The probability distribution of the remaining running time of all particles is then calculated as the remaining lifetime distribution.

[0008] Optionally, the step of mapping different element contents to corresponding wear component sources based on the material composition of different components in the power plant's rotating equipment to construct a wear distribution map includes: Read the content of multiple elements in the oil index data, treat the content of each element as an independent component, and construct an element content vector. Retrieve a pre-stored component material feature library, which contains a material feature vector for each worn component. Each component in the material feature vector corresponds to the theoretical proportion coefficient of an element in the material of that component. The element content vector is multiplied by the material feature vector of each worn component to obtain the similarity response value for each worn component. The similarity response values ​​corresponding to each worn part are sorted by part name to form a response value sequence; A wear distribution map is plotted with the name of the worn component as the horizontal axis label and the similarity response value in the response value sequence as the vertical axis value.

[0009] Optionally, acquiring the oil index data and vibration data associated with the QR code label at the device sampling port includes: Scan the QR code label set at the sampling port of the device, and read the device identifier and sampling time window carried by the QR code label; Within the sampling time window, the lubricating oil sample collected from the sampling port of the device is analyzed by the oil detection terminal to generate initial oil index data; Initial vibration data matching the sampling time window is retrieved from the vibration monitoring system. The initial vibration data is collected by a vibration sensor installed at the bearing housing of the equipment and includes a timestamp. The device identifier is written into the header of the initial oil index data and the initial vibration data, respectively, to obtain the oil index data carrying the device identifier and the vibration data carrying the device identifier.

[0010] Optionally, the step of determining the current wear-dominant component of the device based on the wear distribution map, and extracting the state feature vector corresponding to the wear-dominant component from the wear distribution map, includes: A sliding window difference operation is performed on the similarity response values ​​in the wear distribution map. The second-order difference value of the similarity response values ​​in each sliding window is calculated. The component at the center of the window corresponding to the maximum second-order difference value is determined as the current wear-dominant component of the device. Using the index position of the wear-dominant component in the response value sequence as the center, a local response subsequence of fixed length is extracted; The local response subsequence is fitted with a polynomial to obtain a set of fitting coefficients, and the set of fitting coefficients is arranged from low to high according to a preset polynomial degree to form a state feature vector.

[0011] Optionally, the step of using the real and imaginary parts of the complex state sequence as two parallel input channels to input into a particle filtering algorithm, and having the particle filtering algorithm output the real and imaginary trajectories of each particle, includes: The real part sequence and the imaginary part sequence are separated from the complex state sequence and used as the first input channel and the second input channel, respectively; Initialize a set of particles, each carrying a state vector containing real and imaginary state components, and initialize the state vector of each particle to the initial value of the corresponding input channel; Read the real and imaginary values ​​at each time point in chronological order, and use the current state vector of each particle to predict the real and imaginary values ​​of the particle at the next time point. The real part error is obtained by performing a difference operation between the read real part value and the real part predicted value; the imaginary part error is obtained by performing a difference operation between the read imaginary part value and the imaginary part predicted value. The comprehensive error of each particle is calculated based on the real and imaginary errors of each particle. The weights of each particle are then redistributed based on the comprehensive error, and particles with weights less than a preset weight threshold are resampled. The real state components of each particle after resampling are connected in chronological order to form the real trajectory, and the imaginary state components of each particle are connected in chronological order to form the imaginary trajectory.

[0012] Optionally, comparing the remaining lifetime distribution with the planned maintenance window in the lifecycle management database, and when the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, pushing maintenance suggestions via a mobile terminal and writing the maintenance suggestions into the lifecycle management database associated with the QR code label, includes: The remaining running time of all particles is extracted from the remaining lifetime distribution. Each remaining running time is divided into a low-level interval and a high-level interval according to its numerical value. The upper boundary of the low-level interval is taken as the conservative lifetime estimate, and the lower boundary of the high-level interval is taken as the optimistic lifetime estimate. The planned maintenance window of the device is read from the lifecycle management database. The planned maintenance window contains the time difference between the most recent planned maintenance date and the next planned maintenance date. The time difference is used as the planning interval. Compare the conservative life estimate with the planned interval. If the conservative life estimate is less than the planned interval, calculate the difference between the optimistic life estimate and the planned interval, and multiply the difference by the probability density integral of the remaining life distribution to obtain the maintenance urgency coefficient. Determine whether the maintenance urgency coefficient exceeds a preset threshold. If it does, generate a maintenance recommendation that includes the date corresponding to the conservative lifespan estimate. The maintenance suggestion is pushed to the pre-bound mobile terminal via wireless network, and the maintenance suggestion and the maintenance urgency coefficient are written into the record field associated with the QR code tag in the life cycle management database.

[0013] Secondly, the present invention also provides a power plant equipment full life cycle information management system, including an acquisition module, a mapping module, a determination module, a combination module, and a comparison module.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein the processor is used to execute a computer program stored in the memory to implement the steps of the above-described method for managing the full life cycle information of power plant equipment.

[0015] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the above-described method for managing the full lifecycle information of power plant equipment.

[0016] The beneficial effects of this invention are as follows: First, by linking and collecting oil index data and vibration data through QR code tags and ensuring time synchronization, a unified time benchmark for multi-source monitoring data can be established. Second, based on the material composition of the equipment, the element content in the oil is mapped to different wear parts to construct a wear distribution map, thereby accurately locating the damage degree of each wear part inside the equipment. Third, the wear distribution map identifies the wear-dominant components and extracts the corresponding state feature vectors, achieving a focus from global wear information to key fault characteristics. Furthermore, the state feature vectors are combined with vibration data, and a particle filter algorithm is used to recursively predict the future wear evolution trend of the equipment, outputting the remaining life distribution, making the life prediction more consistent with the actual degradation process of the equipment. Finally, the remaining life distribution is compared with the planned maintenance window, and maintenance suggestions are pushed and written to the database when the remaining life is lower than the maintenance interval, forming a closed-loop management from data collection to maintenance decision-making.

[0017] Furthermore, this invention maps the state feature vector to different phase angles on the complex plane to generate complex feature components, then performs complex multiplication with the real part of the vibration data to construct a complex joint vector, and arranges them in chronological order to form a complex state sequence, achieving deep complex domain fusion of two heterogeneous data sources. Then, using the real and imaginary parts of the complex state sequence as dual-channel inputs to a particle filtering algorithm, it outputs the real and imaginary trajectories respectively. The remaining running time of each particle is calculated through modulus length calculation, and its probability distribution is statistically analyzed as the remaining lifetime distribution. This method preserves the phase relationship between oil wear characteristics and vibration response through complex domain representation, and improves the stability of state tracking and the confidence level of prediction results through the dual-channel parallel processing of particle filtering. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a method for managing the entire lifecycle information of power plant equipment, as provided in an embodiment of the present invention;

[0020] Figure 2 This is a flowchart illustrating the data collection and QR code association process provided in an embodiment of the present invention.

[0021] Figure 3 This is a flowchart illustrating the construction process of a wear distribution map provided in an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the wear distribution pattern provided in an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram illustrating the principle of determining wear-dominant components and extracting state feature vectors according to an embodiment of the present invention.

[0024] Figure 6 This is a schematic diagram illustrating the fusion of complex plane mapping and complex multiplication provided in an embodiment of the present invention;

[0025] Figure 7 This is a flowchart of particle filter dual-channel tracking and remaining lifetime prediction provided in an embodiment of the present invention;

[0026] Figure 8 This is a schematic diagram of the modulus attenuation curve and remaining lifetime determination provided in an embodiment of the present invention;

[0027] Figure 9 A flowchart for maintenance decision comparison and push provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0029] In monitoring the lubrication condition of rotating equipment in power plants, existing technologies typically rely on independent oil analysis or vibration analysis to determine equipment wear. Oil testing has a long sampling cycle and cannot capture sudden faults in real time, while vibration monitoring, although continuously reflecting operating conditions, struggles to identify the specific component source of wear and its remaining lifespan. This separate monitoring approach results in a lack of effective correlation between wear and vibration information, leading to inaccurate judgments of equipment maintenance timing and a mismatch between planned maintenance windows and actual lifespan.

[0030] To address the aforementioned shortcomings, the core idea of ​​this invention is as follows: First, wear chemical information inside the equipment is obtained through oil analysis and converted into quantitative indicators of the wear degree of each component. Then, this chemical information needs to be deeply integrated with vibration physical information to comprehensively utilize the advantages of both types of data for life prediction. Since the physical dimensions and time characteristics of the two types of data are different, they cannot be simply spliced ​​together. Therefore, this invention introduces a complex domain representation method to encode the two types of heterogeneous data into a unified complex space. Then, a particle filtering algorithm is used to track the dynamic evolution of the equipment state in the complex space, ultimately realizing the probability distribution prediction of the remaining life and linking the prediction results with the maintenance plan to form a closed-loop management.

[0031] See Figure 1 As shown, the power plant equipment full lifecycle information management method provided in this embodiment of the invention includes the following steps: Step 101: Obtain oil index data and vibration data associated with the QR code label at the sampling port of the device. The oil index data includes the content of multiple elements in the lubricating oil sample, and the acquisition time of the vibration data is synchronized with the sampling time of the lubricating oil sample.

[0032] The purpose of this step is to establish a spatiotemporal correlation between oil data and vibration data. Since oil analysis reflects wear information accumulated over a period of time, while vibration signals reflect instantaneous operating conditions, meaningful data fusion can only be achieved in subsequent steps if both are collected within the same time window. This invention utilizes QR code tags as a physical bridge to bind device identity, sampling time constraints, and the two types of data together.

[0033] In this embodiment, the equipment sampling port refers to a dedicated interface installed on the equipment's lubrication system pipeline for extracting lubricating oil samples. A QR code label is a graphic encoding identifier capable of storing information, affixed near the equipment sampling port. It pre-codes the equipment's unique identifier and sampling time window. The equipment identifier is a unique code assigned to each device, used to distinguish different devices in the management system. The sampling time window is a relative time period starting from the QR code scanning time; its length is preset by the system. Oil samples collected within this time period are considered to be synchronized with the vibration data time.

[0034] Oil performance data refers to the composition data obtained from the lubricating oil of the equipment, specifically the concentration values ​​of different metal elements such as iron, copper, aluminum, and zinc in the oil sample, expressed in micrograms per milliliter. Vibration data consists of mechanical vibration signals continuously collected by vibration sensors installed at the bearing housing of the equipment. Each vibration data point is accompanied by a timestamp to record the specific time when the data was collected.

[0035] See Figure 2 As shown, the specific implementation process of this step is as follows: First, maintenance personnel scan the QR code label affixed to the sampling port of the equipment using a handheld terminal. After parsing the QR code image, the handheld terminal extracts the pre-stored equipment identifier and sampling time window. By scanning the QR code, maintenance personnel can automatically obtain the equipment identity and sampling time constraints without manually entering equipment information.

[0036] Then, within the time frame specified by the sampling time window, a certain amount of lubricating oil is extracted from the sampling port of the device as a sample, and this sample is injected into the oil detection terminal. The oil detection terminal is a portable analytical instrument that uses spectral analysis technology to irradiate the sample. The working principle of spectral analysis technology is that different metal elements produce characteristic spectral lines of different wavelengths after being excited. By detecting the intensity of each characteristic spectral line, the content of each metal element in the oil can be calculated. The terminal generates a record containing the content values ​​of each of the multiple metal elements, called the initial oil index data.

[0037] Simultaneously, initial vibration data matching the sampling time window is retrieved from the vibration monitoring system. The vibration monitoring system is a platform that continuously receives and stores vibration sensor signals. The vibration sensors are installed at the bearing housings of the equipment to sense the mechanical vibration of the equipment and convert it into electrical signals. The vibration monitoring system filters all vibration data points whose timestamps fall within the sampling time window from the database and packages these data points in chronological order to form the initial vibration data.

[0038] Finally, the equipment identifier is written into the header of both the initial oil level data and the initial vibration data. The header is a reserved field at the beginning of the data record used to store additional information. After the above processing, both the oil level data and the vibration data are marked with the identification information of the same equipment, making it easy to confirm in subsequent steps that the two sets of data belong to the same equipment.

[0039] Step 102: Based on the material composition of different components in the power plant's rotating equipment, map the different element contents to the corresponding wear component sources to construct a wear distribution map, which is used to characterize the wear degree of each wear component inside the equipment.

[0040] After obtaining the oil index data in step 101, the raw element concentration data needs to be converted into component wear information with engineering significance. The raw data only reflects the total concentration of various metal elements in the lubricating oil and cannot directly determine which component is wearing. This step introduces a component material feature library and uses dot product operations to decompose the element concentration signal onto each component, thereby constructing a wear distribution map that can intuitively reflect the wear degree of each component. This is the basis for subsequently identifying the dominant wear component and extracting feature vectors.

[0041] In this embodiment, the rotating equipment in a power plant refers to rotating mechanical equipment that includes wear components such as bearings, gears, and bushings. The wear distribution map is a bar chart data structure with the name of the wear component as the horizontal axis label and the similarity response value of each component as the vertical axis value, where the height of each bar directly represents the wear degree of the corresponding component.

[0042] See Figure 3 As shown, the specific process for constructing the wear distribution map is as follows: The first step is to read the content of multiple elements in the oil index data, treat each element content as an independent component, and construct an element content vector. The element content vector is a one-dimensional array formed by arranging the concentration values ​​of multiple metal elements in a fixed order. For example, if the oil index data shows that the iron content is 8.2 μg / mL, the copper content is 3.5 μg / mL, the aluminum content is 1.8 μg / mL, and the zinc content is 2.1 μg / mL, then the element content vector constructed in the order of iron, copper, aluminum, and zinc is [8.2, 3.5, 1.8, 2.1].

[0043] The second step is to retrieve the pre-stored component material feature library. The component material feature library is a pre-built data table that stores the theoretical proportion coefficients of various metal elements corresponding to each worn component. The material feature vector is a one-dimensional array formed by arranging the theoretical proportion coefficients of each element corresponding to a component in the order of the same element. The theoretical proportion coefficient refers to the dimensionless value obtained by dividing the mass percentage of a certain element in the component material by 100. For example, assuming that iron accounts for 90% and copper accounts for 10% of the bearing material, the material feature vector of the bearing component is [0.9, 0.1, 0.0, 0.0]; assuming that iron accounts for 80% and aluminum accounts for 20% of the gear material, the material feature vector of the gear component is [0.8, 0.0, 0.2, 0.0]; assuming that copper accounts for 70% and zinc accounts for 30% of the bearing bush material, the material feature vector of the bearing bush component is [0.0, 0.7, 0.0, 0.3].

[0044] The third step involves performing a dot product operation between the element content vector and the material feature vector of each worn component to obtain the similarity response value for each component. The dot product operation involves multiplying the values ​​at the same positions in two vectors and then summing all the products to obtain a single scalar value. The similarity response value measures the degree of matching between the element content distribution in the oil and the material characteristics of a particular component; a higher value indicates a more significant contribution to the wear of that component.

[0045] Following the previous numerical example, the similarity response values ​​for each component are calculated as follows: Similarity response values ​​of bearing components: 8.2×0.9+3.5×0.1+1.8×0.0+2.1×0.0=7.38+0.35+0+0=7.73 Similarity response value of gear components: 8.2×0.8+3.5×0.0+1.8×0.2+2.1×0.0=6.56+0+0.36+0=6.92 Similarity response value of bearing components: 8.2×0.0+3.5×0.7+1.8×0.0+2.1×0.3=0+2.45+0+0.63=3.08

[0046] The fourth step is to arrange the similarity response values ​​corresponding to each worn component according to the order of the component's physical installation position in the equipment, forming a response value sequence. This physical installation order ensures that components in adjacent positions in the response value sequence are also spatially adjacent, providing a physically meaningful sorting basis for the sliding window difference operation in the subsequent step 103. Continuing with the previous numerical example, the response value sequence is [7.73, 6.92, 3.08].

[0047] Fifth, plot the wear distribution map with the name of the worn component as the horizontal axis label and the similarity response values ​​in the response value sequence as the vertical axis values. See also Figure 4 As shown in the graph, the taller the bar, the more severe the wear. In this example, the bearing component has a bar height of 7.73, the gear component has a bar height of 6.92, and the bushing component has a bar height of 3.08. It can be clearly seen from the graph that the bearing component has the highest degree of wear.

[0048] This step transforms the concentration data at the element level into wear level data at the component level by performing a dot product operation between the content of multiple elements in the oil and the material feature vectors of each component, thereby achieving precise location and visualization of the wear source.

[0049] Step 103: Determine the current wear-dominant component of the equipment based on the wear distribution map, and extract the state feature vector corresponding to the wear-dominant component from the wear distribution map.

[0050] In step 102, a wear distribution map was constructed, which visually shows the degree of wear on each component. Intuitively, the component with the highest pillar shows the most severe wear. However, in actual engineering, determining the dominant wear component cannot be based solely on the absolute value; abrupt changes in wear degree between adjacent components must also be considered. This is because the dominant wear component is often not the component with the highest wear value, but rather the component whose wear degree changes most drastically relative to its surrounding components. For example, if the wear values ​​of all components are uniformly high, it indicates overall aging rather than localized deterioration of a single component; while if the wear value of a component changes sharply relative to its left and right neighboring components, it indicates that the component is experiencing localized accelerated wear and requires close attention. Therefore, this step uses a sliding window second-order difference operation to identify the location of the most drastic change in wear degree and determines the corresponding component as the dominant wear component.

[0051] Furthermore, in order to pass the wear state information of the wear-dominant component to the subsequent step 104 for life prediction, it is not sufficient to simply pass a scalar similarity response value. Instead, a set of feature parameters that can describe the wear distribution pattern around the component needs to be extracted. This step achieves this goal by performing polynomial fitting on the local response values ​​around the wear-dominant component: the coefficients of each order obtained from the fitting describe the baseline level, trend of change, and curvature of the local wear distribution, respectively. These parameters form a state feature vector, providing a structured feature input for subsequent fusion with vibration data.

[0052] See Figure 5 As shown, the specific process for determining the wear-prone component is as follows: The wear-dominant component is the component with the most dramatic change in wear degree among all wear components. To locate this component, a sliding window differencing operation is performed on the similarity response values ​​in the wear distribution map. The sliding window differencing operation involves sliding a fixed-length window along the response value sequence, calculating the second-order difference value of the sequence values ​​at each window position. The second-order difference value measures the degree of curvature change of the sequence within that window, i.e., the change in the rate of change between adjacent values. The larger the absolute value of the second-order difference, the more dramatic the abrupt change in wear degree at the center of the window. The component corresponding to the center position of the window with the largest absolute value of the second-order difference is identified as the wear-dominant component.

[0053] The sliding window length is set to an odd number to ensure that each window has a unique center position. Continuing with the previous numerical example, the response value sequence is [7.73, 6.92, 3.08], and the window length is set to 3. Since the sequence length is exactly 3, only one window covers all three components, and the center position of this window corresponds to the gear component with index 2. The difference value of this window is calculated as follows:

[0054] First-order difference:

[0055]

[0056] Second-order difference:

[0057] The absolute value of the second difference is |-3.03| = 3.03. In this example, there is only one window, so the absolute value of the second difference of this window, 3.03, is the maximum value. The center position of the window corresponding to this maximum absolute value of the second difference is the gear component, so the gear component is identified as the current dominant wear component. Physically speaking, the wear value at the gear component decreases by only 0.81 from its left neighbor (bearing 7.73) to itself (6.92), but drops sharply by 3.84 from itself to its right neighbor (bearing bush 3.08). This sharp increase in the rate of change (i.e., the large absolute value of the second difference) indicates that the gear component is at the steepest point of the wear gradient and is the key component where the wear state changes significantly.

[0058] In practical engineering applications, when the equipment contains a larger number of wear parts (e.g., more than 10), the sliding window will move step by step along the sequence, calculating the second-order difference at each position, and finally selecting the position with the most drastic change. If the absolute values ​​of the second-order differences at multiple positions are close, they can all be considered as candidate dominant wear parts for subsequent analysis.

[0059] See Figure 5 As shown, the specific process for extracting the state feature vector is as follows: After identifying the dominant wear component, a set of numerical features describing the wear distribution morphology around that component needs to be extracted from the wear distribution map. These features are then passed to the subsequent step 104 as input to the particle filtering algorithm. If only the similarity response value (a scalar) of the dominant wear component itself is passed, the trend and curvature information of the wear distribution around that component is lost. To preserve richer morphological information, this step uses the index position of the dominant wear component in the response value sequence as the center, extracts a fixed-length local response subsequence, and then performs polynomial fitting on this subsequence, using the fitting coefficients to compactly characterize the morphology of the local wear distribution.

[0060] Specifically, taking the index position of the wear-dominant component in the response value sequence as the center, a predetermined number of positions are extended to the left and right to extract a local response subsequence. When the extension boundary exceeds the start or end position of the response value sequence, boundary value filling is used for processing. Continuing with the previous numerical example, the index position of the gear component is 2, the extraction length is 3, and the resulting local response subsequence is [7.73, 6.92, 3.08].

[0061] Then, a polynomial fitting is performed on the local response subsequence. Polynomial fitting is a numerical method that uses a polynomial function to approximate the distribution shape of data points. The polynomial degree is set to 2, meaning a polynomial of the form shown is fitted. The quadratic function is used. Three data points in the subsequence are numbered according to their positions in the sequence (e.g., x=1,2,3) as independent variables, and the similarity response value is used as the dependent variable. The fitting coefficients a, b, and c are solved using the least squares method. The physical meaning of the fitting coefficients is as follows: the constant term c reflects the average baseline level of the local wear distribution; the linear coefficient b reflects the linear trend of the wear degree along the component arrangement direction (positive values ​​indicate increasing wear, negative values ​​indicate decreasing wear); and the quadratic coefficient a reflects the curvature of the wear distribution (i.e., the acceleration or deceleration of the trend).

[0062] The fitting coefficients are arranged in ascending order of polynomial degree, i.e., [c, b, a], to form a state feature vector. This state feature vector compactly describes the wear distribution pattern around the wear-dominant component with three values, providing structured input features for the complex domain fusion in the subsequent step 104.

[0063] Step 104: Combine the state feature vector with the vibration data, and based on the correspondence between the sampling time and the acquisition time, use the particle filter algorithm to perform state tracking on the combined data to recursively predict the wear state evolution trend of the equipment in future operating cycles and output the remaining life distribution.

[0064] In the preceding steps, steps 102-103 extracted state feature vectors representing the wear distribution pattern from the oil data, and step 101 acquired time-synchronized vibration data. The core task of this step is to deeply fuse these two types of data from different physical domains, and use the fused data to track the evolution of the equipment state over time, ultimately predicting the remaining lifespan of the equipment.

[0065] The key challenge here is that the state feature vector is a set of fitting coefficients describing the wear distribution pattern (chemical domain information), while the vibration amplitude is a physical quantity reflecting the instantaneous mechanical operating state (physical domain information). Their physical dimensions and numerical characteristics are completely different, and they cannot be directly concatenated or simply added. To solve this problem, this invention introduces a complex domain representation method: each component of the state feature vector is encoded as a component in a complex plane with different phase angles, and then the vibration amplitude is fused with these components through complex multiplication. The advantage of this approach is that complex multiplication inherently possesses the mathematical properties of amplitude modulation and phase superposition, enabling the simultaneous capture of the correlation strength (through amplitude) and temporal sequence (through phase) of the two types of data in a single calculation, avoiding the information loss caused by simple concatenation.

[0066] The fused complex state sequence changes over time, forming a trajectory moving on the complex plane. The modulus of this trajectory (i.e., the absolute value of the complex number) comprehensively reflects the overall health status of the equipment: when the equipment is in good condition, vibration energy is stable, wear is low, and the modulus remains at a high level; as wear intensifies and vibration abnormalities worsen, the modulus gradually decays; when the modulus drops below a preset end-of-life threshold, the equipment is considered to have reached the end of its service life. Therefore, tracking the decay trend of the modulus can predict the remaining lifespan.

[0067] Due to the noise and uncertainties inherent in real-world systems, directly extrapolating the modulus length can lead to significant deviations. Therefore, this invention employs a particle filtering algorithm for state tracking. The particle filtering algorithm is a sequential state estimation method based on Monte Carlo sampling. Its core idea is to use a set of weighted random particles to approximate the probability distribution of the system state. Each particle independently predicts the future state, adjusting its weight by comparing it with actual observations; particles with higher weights represent more likely state trajectories. Finally, by statistically analyzing the remaining running time of all particles, the probability distribution of the remaining lifetime can be obtained, rather than a single point estimate. This provides a more comprehensive risk assessment basis for maintenance decisions.

[0068] The following is a detailed step-by-step explanation of the specific implementation process of this step: Step 1: Map the state feature vectors to the complex plane (see...) Figure 6 ).

[0069] The state feature vector contains multiple fitting coefficients, each reflecting an aspect of the wear distribution pattern. To encode these coefficients into the complex space, a fixed phase angle is pre-assigned to each fitting coefficient, allowing coefficients of different orders to occupy different directions on the complex plane. The mapping rule is as follows: using the numerical value of the fitting coefficient as the modulus and the preset phase angle θ as the direction angle, the corresponding complex value is calculated using Euler's formula. z = r × (cosθ + i × sinθ);

[0070] Where r is the value of the fitting coefficient, θ is the preset phase angle corresponding to the coefficient, and i is the imaginary unit.

[0071] For example, the state feature vector is [7.73, 2.46, -1.23] (the coefficients of the constant term, the first term, and the second term, respectively), and the three fitting coefficients are preset with phase angles of 0°, 60°, and 120°, respectively.

[0072] The calculation is as follows: The constant term coefficient 7.73 maps to θ=0°:

[0073] The coefficient of the linear term, 2.46, maps to θ = 60°:

[0074] The quadratic coefficient -1.23 maps to θ = 120°:

[0075] Through the above mapping, the three dimensions of wear distribution pattern are encoded into components in three different directions on the complex plane, which prepares for subsequent complex multiplication fusion with vibration data.

[0076] Step 2: Perform complex multiplication on the vibration amplitude and the complex characteristic components to obtain the complex joint vector (see...). Figure 6 ).

[0077] For each sampling time point, the vibration amplitude v at that time point is treated as a pure real number and multiplied by the corresponding complex characteristic component z. Since the vibration amplitude is a pure real number, the complex multiplication simplifies to: w = v × z = v × (a + bi) = va + vbi; Here, a and b are the real and imaginary parts of the complex characteristic components, respectively. The physical meaning of the multiplication result w is as follows: the real part va represents the projection of vibration energy onto the in-phase direction of the wear characteristic, reflecting the temporal synchronous correlation strength between vibration and wear; the imaginary part vb represents the projection of vibration energy onto the orthogonal direction of the wear characteristic, reflecting the phase difference or hysteresis relationship between the two. Together, they constitute a complex joint vector that integrates vibration physical information and wear chemical information.

[0078] For example, suppose the vibration amplitude at a certain time point is v=5.2, and the corresponding complex characteristic component is... Then the joint vector of complex numbers is: w = 5.2 × (7.73 + 0i) = 40.196 + 0i;

[0079] Step 3: Construct a complex state sequence.

[0080] The complex joint vectors at each time point are arranged in ascending order to construct a complex state sequence. This sequence serves as the input data for the subsequent particle filter algorithm. For example, assuming there are three time points with corresponding complex joint vectors of 40.196+0i, 35.724+2.456i, and 28.956+3.892i, the complex state sequence would be [40.196+0i, 35.724+2.456i, 28.956+3.892i]. It can be observed that the magnitude of the complex joint vector decreases over time, reflecting the process of increased equipment wear and tear and a decline in equipment health.

[0081] Step 4: Dual-channel particle filter tracking (see...) Figure 7 ).

[0082] The real and imaginary parts of the complex state sequence are separated and used as two parallel input channels to feed into the particle filter algorithm. The reason for using dual channels instead of a single channel is that the real and imaginary parts carry information from different dimensions (co-correlation and orthogonal correlation) in the fused data. Processing them separately can avoid information aliasing, and at the same time, the synergistic relationship between the two channels can be used to improve tracking accuracy.

[0083] The specific process is as follows: First, a set of particles is initialized, each carrying a state vector containing real and imaginary state components, initialized to the initial values ​​of the corresponding channel. Then, the process proceeds sequentially over time: at each time step, each particle uses a state transition function to predict the state value at the next time step. The state transition function is a mathematical function describing the evolution of the system from the current state to the next state. After the prediction is completed, the predicted value is compared with the actual observed value, and the combined error is calculated. ;

[0084] in, The real part error is the difference between the actual real part value and the predicted real part value. The imaginary part error is calculated as the actual imaginary part value minus the predicted imaginary part value. Particles with smaller overall errors indicate that their predictions are closer to the true state, and therefore are assigned larger weights. Particles with weights less than a preset threshold are discarded, while particles with larger weights are duplicated to fill gaps; this operation is called resampling. After resampling, the particle swarm concentrates in the more probable state region, achieving effective tracking of the device state.

[0085] The real part of the state components of each particle at each time step is connected in time to form the real part trajectory, and the imaginary part of the state components is connected in time to form the imaginary part trajectory.

[0086] Step 5: Calculation of module length and output of remaining lifetime distribution (see...) Figure 8 ).

[0087] The modulus of the trajectory is calculated for both the real and imaginary parts of each particle's trajectory to obtain the particle's modulus trajectory. The formula for calculating the modulus is: ;

[0088] Where x(t) is the real part of the state component at time t, and y(t) is the imaginary part of the state component at time t. The modulus length comprehensively reflects the overall strength of the fused data at that time: the modulus length is large when the data is in good health, and gradually decreases as wear intensifies.

[0089] See Figure 8 As shown, for each particle, starting from the current moment of the modulus trajectory and traversing backwards (i.e., extrapolating to future time periods using the state transition function), the time point when the modulus value first decays below the preset lifetime threshold is found. The time difference between this time point and the current moment is taken as the particle's remaining running time. Due to differences in random perturbations, different particles have different predicted remaining running times. The remaining running times of all particles are statistically analyzed to generate a probability distribution histogram, which is the remaining lifetime distribution. Compared to a single point estimate, the remaining lifetime distribution can simultaneously provide the most likely lifetime value and the range of uncertainty, providing a more comprehensive risk assessment basis for subsequent maintenance decisions.

[0090] Step 105: Compare the remaining lifetime distribution with the planned maintenance window in the lifecycle management database. When the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, push maintenance suggestions through the mobile terminal and write the maintenance suggestions into the lifecycle management database associated with the QR code label.

[0091] In step 104, the remaining lifetime distribution of the equipment was obtained. However, the remaining lifetime distribution alone cannot directly guide maintenance actions; it needs to be compared with the equipment's existing planned maintenance schedule to determine whether the equipment can safely operate until the next planned maintenance time. If the predicted remaining lifetime is shorter than the planned maintenance interval, it indicates that the equipment may fail before the planned maintenance, requiring maintenance to be scheduled in advance. This step transforms the remaining lifetime distribution into actionable maintenance recommendations, completing the final closed loop from condition prediction to maintenance decision-making.

[0092] In this embodiment, the lifecycle management database is a storage system for the entire lifecycle data of the device, including basic device information, maintenance history, and planned maintenance windows. The planned maintenance window contains the time difference between the most recent planned maintenance date and the next planned maintenance date. The mobile terminal is a smart device carried by maintenance personnel.

[0093] See Figure 9 As shown, the specific implementation process of this step is as follows: First, extract the remaining runtime of all particles from the remaining lifetime distribution and sort them in ascending order. Divide the sorted sequence into two halves, the first half forming the lower-order interval and the second half forming the higher-order interval. Take the upper boundary of the lower-order interval as the conservative lifetime estimate T. l Take the lower boundary of the high-level interval as the optimistic lifetime estimate T. h Conservative life estimates represent the remaining life of the equipment under a more pessimistic scenario, while optimistic life estimates represent the remaining life under a more optimistic scenario. For example, assuming the remaining operating times for the five particles are 10 days, 12 days, 15 days, 18 days, and 20 days, then... sky, sky.

[0094] Then, the planned maintenance window of the equipment is read from the lifecycle management database, and the planned interval is calculated. For example, if the most recent planned maintenance date is March 1, 2025, and the next one is June 1, 2025, then... sky.

[0095] Next, compare the conservative lifespan estimate T. l Interval with plan .like This indicates that the equipment may fail before the planned maintenance, and it is necessary to calculate the maintenance urgency factor K: ;

[0096] Where ∫f(t)dt is the probability density integral of the remaining lifetime distribution, i.e., the sum of the weights of all particles. The maintenance urgency coefficient K is a dimensionless value used to quantify the urgency of maintenance adjustments.

[0097] Furthermore, determine whether K exceeds a preset coefficient threshold. .like Then add the conservative life estimate T to the current date. l Given the target date, generate maintenance recommendations that include that date. For example, if the current date is March 15, 2025... If the problem persists, it is recommended that the overhaul be completed by March 27, 2025.

[0098] Finally, the maintenance recommendations are pushed to the pre-bound mobile terminal via wireless network and simultaneously written to the record field associated with the QR code label in the lifecycle management database to complete the persistent storage of the data.

[0099] This invention also provides a power plant equipment lifecycle information management system, comprising: The acquisition module is used to acquire oil index data and vibration data associated with the QR code label at the sampling port of the equipment; The mapping module is used to map the element content to the corresponding wear component source based on the material composition of different components in the rotating equipment of the power plant, so as to construct a wear distribution map. The determination module is used to identify the wear-dominant components based on the wear distribution map and extract the corresponding state feature vectors; The joint module is used to combine the state feature vector with the vibration data in the complex domain, use the particle filter algorithm for state tracking, and output the remaining lifetime distribution. The comparison module is used to compare the remaining life distribution with the planned maintenance window, push maintenance suggestions, and write them to the lifecycle management database.

[0100] The specific implementation of the above system can be referred to the description of the method embodiments above, and will not be repeated here.

[0101] The present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the power plant equipment lifecycle information management method described above.

[0102] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described power plant equipment lifecycle information management methods.

[0103] The technical solution provided by this invention has been described in detail above. For those skilled in the art, various improvements and modifications can be made to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of this invention.

Claims

1. A method for managing the entire lifecycle information of power plant equipment, characterized in that, include: The oil index data and vibration data associated with the QR code label at the sampling port of the device are obtained. The oil index data includes the content of multiple elements in the lubricating oil sample, and the acquisition time of the vibration data is synchronized with the sampling time of the lubricating oil sample. Based on the material composition of different components in the rotating equipment of the power plant, the different contents of the aforementioned elements are mapped to the corresponding wear component sources to construct a wear distribution map, which is used to characterize the wear degree of each wear component inside the equipment; The wear-dominant component of the equipment is determined based on the wear distribution map, and the state feature vector corresponding to the wear-dominant component is extracted from the wear distribution map. The state feature vector is combined with the vibration data. Based on the correspondence between the sampling time and the acquisition time, the particle filter algorithm is used to perform state tracking on the combined data to recursively predict the wear state evolution trend of the equipment in future operating cycles and output the remaining life distribution. The remaining lifetime distribution is compared with the planned maintenance window in the lifecycle management database. When the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, a maintenance suggestion is pushed through the mobile terminal and written into the lifecycle management database associated with the QR code label.

2. The method according to claim 1, characterized in that, The step of combining the state feature vector with the vibration data, and using a particle filter algorithm to perform state tracking on the combined data based on the correspondence between the sampling time and the acquisition time, to recursively predict the wear state evolution trend of the equipment in future operating cycles, and output the remaining lifetime distribution, includes: Each fitting coefficient in the state feature vector is mapped to a different phase angle in the complex plane to generate a set of complex feature components; Each vibration amplitude in the vibration data is taken as the real part and multiplied by the complex feature component at the corresponding sampling time point to obtain a complex joint vector; Using the correspondence between the sampling time and the acquisition time as the time axis, the complex joint vectors at each time point are arranged in ascending order of time to construct a complex state sequence; The real and imaginary parts of the complex state sequence are used as two parallel input channels and input into the particle filtering algorithm. The particle filtering algorithm outputs the real and imaginary trajectories of each particle. The modulus of the real and imaginary trajectories of each particle is calculated to obtain the remaining running time for each particle. The probability distribution of the remaining running time of all particles is then calculated as the remaining lifetime distribution.

3. The method according to claim 1, characterized in that, Based on the material composition of different components in the power plant's rotating equipment, different element contents are mapped to corresponding wear component sources to construct a wear distribution map. This wear distribution map characterizes the wear degree of each wear component within the equipment, including: Read the content of multiple elements in the oil index data, treat the content of each element as an independent component, and construct an element content vector. Retrieve a pre-stored component material feature library, which contains a material feature vector for each worn component. Each component in the material feature vector corresponds to the theoretical proportion coefficient of an element in the material of that component. The element content vector is multiplied by the material feature vector of each worn component to obtain the similarity response value for each worn component. The similarity response values ​​corresponding to each worn part are sorted by part name to form a response value sequence; A wear distribution map is plotted with the name of the worn component as the horizontal axis label and the similarity response value in the response value sequence as the vertical axis value.

4. The method according to claim 1, characterized in that, The acquisition of oil index data and vibration data associated with the QR code label at the sampling port of the equipment includes: Scan the QR code label set at the sampling port of the device, and read the device identifier and sampling time window carried by the QR code label; Within the sampling time window, the lubricating oil sample collected from the sampling port of the device is analyzed by the oil detection terminal to generate initial oil index data; Initial vibration data matching the sampling time window is retrieved from the vibration monitoring system. The initial vibration data is collected by a vibration sensor installed at the bearing housing of the equipment and includes a timestamp. The device identifier is written into the header of the initial oil index data and the initial vibration data, respectively, to obtain the oil index data carrying the device identifier and the vibration data carrying the device identifier.

5. The method according to claim 1, characterized in that, The step of determining the current wear-dominant component of the equipment based on the wear distribution map, and extracting the state feature vector corresponding to the wear-dominant component from the wear distribution map, includes: A sliding window difference operation is performed on the similarity response values ​​in the wear distribution map. The second-order difference value of the similarity response values ​​in each sliding window is calculated. The component at the center of the window corresponding to the maximum second-order difference value is determined as the current wear-dominant component of the device. Using the index position of the wear-dominant component in the response value sequence as the center, a local response subsequence of fixed length is extracted; The local response subsequence is fitted with a polynomial to obtain a set of fitting coefficients, and the set of fitting coefficients is arranged from low to high according to a preset polynomial degree to form a state feature vector.

6. The method according to claim 2, characterized in that, The step of using the real and imaginary parts of the complex state sequence as two parallel input channels to input into a particle filtering algorithm, and having the particle filtering algorithm output the real and imaginary trajectories of each particle, includes: The real part sequence and the imaginary part sequence are separated from the complex state sequence and used as the first input channel and the second input channel, respectively; Initialize a set of particles, each carrying a state vector containing real and imaginary state components, and initialize the state vector of each particle to the initial value of the corresponding input channel; Read the real and imaginary values ​​at each time point in chronological order, and use the current state vector of each particle to predict the real and imaginary values ​​of the particle at the next time point. The real part error is obtained by performing a difference operation between the read real part value and the real part predicted value; the imaginary part error is obtained by performing a difference operation between the read imaginary part value and the imaginary part predicted value. The comprehensive error of each particle is calculated based on the real and imaginary errors of each particle. The weights of each particle are then redistributed based on the comprehensive error, and particles with weights less than a preset weight threshold are resampled. The real state components of each particle after resampling are connected in chronological order to form the real trajectory, and the imaginary state components of each particle are connected in chronological order to form the imaginary trajectory.

7. The method according to claim 1, characterized in that, The step of comparing the remaining lifetime distribution with the planned maintenance window in the lifecycle management database, and when the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, pushing maintenance suggestions via a mobile terminal and writing the maintenance suggestions into the lifecycle management database associated with the QR code label, includes: The remaining running time of all particles is extracted from the remaining lifetime distribution. Each remaining running time is divided into a low-level interval and a high-level interval according to its numerical value. The upper boundary of the low-level interval is taken as the conservative lifetime estimate, and the lower boundary of the high-level interval is taken as the optimistic lifetime estimate. The planned maintenance window of the device is read from the lifecycle management database. The planned maintenance window contains the time difference between the most recent planned maintenance date and the next planned maintenance date. The time difference is used as the planning interval. Compare the conservative life estimate with the planned interval. If the conservative life estimate is less than the planned interval, calculate the difference between the optimistic life estimate and the planned interval, and multiply the difference by the probability density integral of the remaining life distribution to obtain the maintenance urgency coefficient. Determine whether the maintenance urgency coefficient exceeds a preset threshold. If it does, generate a maintenance recommendation that includes the date corresponding to the conservative lifespan estimate. The maintenance suggestion is pushed to the pre-bound mobile terminal via wireless network, and the maintenance suggestion and the maintenance urgency coefficient are written into the record field associated with the QR code tag in the life cycle management database.

8. A power plant equipment full lifecycle information management system, characterized in that, include: The acquisition module is used to acquire oil index data and vibration data associated with the QR code label at the sampling port of the device. The oil index data includes the content of multiple elements in the lubricating oil sample, and the acquisition time of the vibration data is synchronized with the sampling time of the lubricating oil sample. The mapping module is used to map different element contents to corresponding wear parts sources based on the material composition of different components in the rotating equipment of the power plant, so as to construct a wear distribution map. The wear distribution map is used to characterize the wear degree of each wear part inside the equipment. The determination module is used to determine the current wear-dominant component of the equipment based on the wear distribution map, and to extract the state feature vector corresponding to the wear-dominant component from the wear distribution map. The joint module is used to combine the state feature vector with the vibration data, and based on the correspondence between the sampling time and the acquisition time, use the particle filter algorithm to perform state tracking on the combined data to recursively predict the wear state evolution trend of the equipment in future operating cycles and output the remaining life distribution. The comparison module is used to compare the remaining lifetime distribution with the planned maintenance window in the life cycle management database. When the remaining lifetime indicated by the remaining lifetime distribution is lower than the next planned maintenance interval in the planned maintenance window, a maintenance suggestion is pushed through the mobile terminal and the maintenance suggestion is written into the life cycle management database associated with the QR code label.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the power plant equipment lifecycle information management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the power plant equipment lifecycle information management method as described in any one of claims 1 to 7.

Images