Centrifugal oil mist generator fault code data processing method and system

Abstract

This invention relates to the field of fault code data processing technology, and particularly to a method and system for processing fault code data of centrifugal oil mist lubricators. The method includes parallel acquisition of operating environment parameters and original fault codes, appending high-precision timestamps to generate a synchronous data stream, calling a fault knowledge base and associating contextual data to form a dataset, extracting time-domain and frequency-domain features to generate multi-dimensional fault feature vectors, encapsulating them into standardized fault case units, storing them for updating the diagnostic prediction model, and optimizing the knowledge base based on maintenance feedback. The system comprises five collaborative modules: synchronous data acquisition, context-aware associative feature adaptive extraction, standardized encapsulation, case storage, and model iteration. This invention achieves standardized and precise data processing, improves fault diagnosis efficiency and accuracy, supports system adaptive evolution, reduces operation and maintenance costs, and is suitable for efficient fault handling of centrifugal oil mist lubricators.

Description

Technical Field

[0001] This invention relates to the field of fault code data processing technology, specifically to a method and system for processing fault code data of centrifugal oil mist lubricators. Background Technology

[0002] Currently, there are many shortcomings in the industry's fault data processing for centrifugal oil mist lubricators. In the data acquisition phase, most systems only acquire the raw fault codes reported by the controller, lacking synchronous acquisition of equipment operating status parameters and environmental parameters, resulting in a lack of comprehensive data support for fault analysis. While some systems attempt to collect data from multiple sources, the acquisition frequencies of different data streams are inconsistent and lack unified time stamping, leading to poor data synchronization and making it difficult to establish accurate correlations between fault events and changes in related parameters.

[0003] In terms of feature extraction, existing technologies mostly rely on manual experience or single-dimensional feature analysis, failing to combine the characteristics of fault modes to extract features from multiple dimensions in the time and frequency domains. This results in incomplete feature representation and difficulty in accurately depicting the essence of the fault. At the same time, the processing of fault data lacks standardized procedures, and the storage formats of relevant information for different fault events are inconsistent, forming data silos and hindering efficient retrieval and reuse of fault cases.

[0004] Furthermore, existing systems often use fixed configurations for fault diagnosis models, lacking incremental update mechanisms based on accumulated real-world fault cases, and have not established an effective maintenance feedback loop. The feature template association rules and other content in the fault knowledge base cannot adaptively adjust to changes in actual application scenarios. This results in difficulty in continuously improving diagnostic accuracy as equipment operating time increases, and even leads to problems such as false positives and false negatives, severely impacting fault handling efficiency and increasing equipment maintenance costs.

[0005] In summary, existing methods for processing fault code data in centrifugal oil mist generators have significant shortcomings in areas such as data acquisition, synchronous feature extraction, case encapsulation, model iteration, and knowledge base evolution. There is an urgent need for a systematic and standardized processing method and system to address issues such as the one-sidedness, insufficient synchronization, low standardization, and lack of adaptive capabilities in data processing, thereby providing a reliable guarantee for accurate fault diagnosis. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for processing fault code data of centrifugal oil mist generators, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The method for processing fault code data of centrifugal oil mist lubricators includes the following steps:

[0009] Step S1: Construct and synchronize data acquisition streams: acquire the operating status time-series data stream of the centrifugal oil mist generator, the environmental parameter time-series data stream, and the original fault code sequence triggered by the event in parallel, and uniformly add high-precision timestamps.

[0010] Step S2: Context-aware data association based on fault knowledge base: In response to the capture of an original fault code, a preset fault knowledge base is invoked. Based on the potential fault mode indicated by the fault code, an association time window and a set of key association parameters are adaptively determined. Data fragments of key association parameters within the time window are extracted from the time-series data stream of operating status and environmental parameters to form an original context dataset strongly correlated with the original fault code.

[0011] Step S3, Adaptive feature extraction and vectorization for diagnosis: The original context dataset is preprocessed, and a multidimensional fault feature vector is automatically calculated and generated from the preprocessed data based on the fault feature template defined for the fault mode in the fault knowledge base; the feature vector includes at least the feature quantities extracted from the time domain, frequency domain, or time-frequency domain to distinguish the fault mode;

[0012] Step S4: Generate standardized and scalable fault case units: Encapsulate the original fault code, multi-dimensional fault feature vector, corresponding fault timestamp, standardized fault identifier obtained from the fault knowledge base, and initial severity assessment into a structured fault case data unit.

[0013] Step S5, Fault Case Storage and Model Iterative Update: Store fault case data units in the case library and use them to update or train the fault diagnosis and prediction model; at the same time, based on the difference between the model diagnosis results and subsequent actual maintenance feedback, generate optimization feedback signals for feature templates or association rules in the fault knowledge base to achieve adaptive evolution of the knowledge base.

[0014] As a preferred approach, the construction and synchronization of the data acquisition stream in step S1 includes:

[0015] The operating status time-series data stream of the centrifugal oil mist generator is collected in real time through hard-wiring, including motor current, motor voltage, fan speed, and inlet / outlet pressure difference. Meanwhile, the environmental parameters time-series data stream of ambient temperature, ambient humidity, and ambient particulate matter concentration are collected asynchronously through a sensor network.

[0016] Listen to the event messages actively reported by the centrifugal oil mist generator controller through the communication bus. When the message type is a fault alarm, parse out the original fault code sequence that triggered the event.

[0017] For each data record of the running status timing data stream, the environmental parameter timing data stream, and the original fault code sequence, a microsecond-level timestamp originating from the same high-precision hardware clock is added;

[0018] The system caches and preprocesses the timestamped multi-channel data, removes obviously invalid and abnormal data points, and aligns the time-series data streams of operating status and environmental parameters at different acquisition frequencies with the time-series data streams based on the timestamps, outputting a cleaned synchronous data acquisition stream.

[0019] As a preferred approach, step S2, which involves context-aware data association based on a fault knowledge base, specifically includes the following steps:

[0020] In response to the capture of a raw fault code, immediately record the fault timestamp corresponding to the event;

[0021] Using the original fault code as an index, query the preset fault knowledge base to obtain the standardized fault identifier bound to the fault code, the preset associated time window configuration, and a set of predefined key associated parameters.

[0022] Based on the fault timestamp and the queried time window configuration, the start and end times for extracting context data are calculated, and the associated time window is defined.

[0023] Based on the list of key correlation parameters, extract time-series data segments corresponding to all key correlation parameters within the correlation time window from the synchronous data acquisition stream aligned with the time axis;

[0024] The extracted time-series data fragments of each key correlation parameter are combined and encapsulated with the original fault code, fault timestamp, and standardized fault identifier to generate an original context dataset strongly correlated with the original fault code.

[0025] As a preferred approach, step S3, the adaptive feature extraction and vectorization for diagnosis, specifically includes the following steps:

[0026] The original context dataset is preprocessed to generate a preprocessed, standardized data sequence. Data preprocessing includes outlier removal, noise filtering, and data normalization.

[0027] Using standardized fault identifiers as indexes, the fault knowledge base is queried to obtain the fault feature template corresponding to the fault mode; the fault feature template defines the types of features to be extracted from the time domain, frequency domain, and time-frequency domain, as well as the calculation methods;

[0028] Based on the fault feature template, feature calculation operations are performed on the preprocessed standardized data sequence to automatically extract time-domain features, frequency-domain features, and time-frequency-domain features.

[0029] The extracted time-domain features, frequency-domain features, and time-frequency-domain features are combined in a preset order to generate a multidimensional fault feature vector.

[0030] As a preferred embodiment, step S4 specifically includes:

[0031] In response to the multidimensional fault feature vector, the fault knowledge base is queried using the standardized fault identifier as an index to obtain the standard fault mode description and formatted storage template corresponding to the standardized fault identifier.

[0032] Based on the standard fault mode description, the fault events indicated by the original fault codes are pattern matched and confirmed, and an initial severity assessment is calculated and generated based on the comparison results of the feature values ​​in the multidimensional fault feature vector with the preset threshold.

[0033] Associate and integrate fault timestamps, standardized fault identifiers, confirmed fault modes, initial severity assessments, and multidimensional fault feature vectors, and attach start and end time markers for the associated time windows;

[0034] Based on the formatted storage template, the integrated information, including the original fault codes, is encapsulated into fault case data units with a unified field structure. The fault case data units are configured to support time-based indexing and fault feature-based retrieval.

[0035] As a preferred approach, step S5, which involves storing fault cases and iteratively updating the model, specifically includes the following steps:

[0036] Fault case data units are written into the fault case library of the distributed time series database, and a globally unique case identifier is assigned to each fault case data unit to establish a fast retrieval structure with fault timestamp and standardized fault identifier as a joint index.

[0037] In response to an increase in the number of cases in the fault case library or a change in the set of case identifiers, the incremental update process of the fault diagnosis and prediction model is triggered. The incremental update process takes all or recently added fault case data units in the case library and their contained multidimensional fault feature vectors as input samples.

[0038] Based on the input sample set, the fault diagnosis and prediction model is trained or retrained to generate or update the model's internal parameters; the performance of the updated model is verified using the reserved validation case set, and the diagnostic accuracy and confusion matrix of the model for each standardized fault identifier are calculated.

[0039] Receive feedback information from actual repair work orders for recorded fault cases. The feedback information includes at least the confirmed actual fault cause, repair measures, and repair verification results. Associate and bind the feedback information with the corresponding fault case data unit in the fault case library through the case identifier.

[0040] The initial diagnostic results of the fault diagnosis and prediction model for a specific fault case are compared with the actual fault causes confirmed in the subsequent actual maintenance feedback information. If there is a difference between the two, the difference analysis process is initiated to generate an optimization feedback signal for the fault knowledge base. The optimization feedback signal includes at least the correction suggestions for the fault feature templates associated with specific standardized fault identifiers, or the adjustment suggestions for the list of key associated parameters and the configuration of associated time windows.

[0041] Based on the optimization feedback signals, after manual review or approval by preset rules, the corresponding entries in the fault knowledge base are updated, including modifying the fault feature template, adjusting the list of key related parameters, or optimizing the configuration of the related time window, so as to achieve the adaptive evolution of the fault knowledge base based on case accumulation and feedback.

[0042] The centrifugal oil mist lubricator fault code data processing system includes:

[0043] The data synchronization acquisition module is configured to acquire the time-series data streams of the centrifugal oil mist generator's operating status and environmental parameters in parallel, and to listen to the event messages reported by the controller to parse the original fault code sequence. It also adds a unified high-precision timestamp to all data, performs time axis alignment and cleaning on multiple data streams, and outputs a synchronous data acquisition stream.

[0044] The context-aware association module is communication-coupled to the data synchronization acquisition module and a fault knowledge base. It is configured to respond to receiving the original fault code and its corresponding timestamp from the data synchronization acquisition module, query the fault knowledge base to obtain the association time window configuration and key association parameter list, define the time window based on the timestamp and the association time window configuration, and extract the time-series data fragments of key association parameters within the time window from the synchronous data acquisition stream to generate the original context dataset corresponding to the original fault code.

[0045] The feature adaptive extraction module is communicatively coupled to the context-aware association module and the fault knowledge base. It is configured to receive the original context dataset, preprocess it to generate a standardized data sequence, and automatically calculate and extract time-domain, frequency-domain and time-frequency-domain feature quantities from the standardized data sequence based on the fault feature template obtained from the fault knowledge base that corresponds to the original fault code, and combine them to generate a multi-dimensional fault feature vector.

[0046] The standardized encapsulation module is communication-coupled to the feature adaptive extraction module, the context-aware association module, and the fault knowledge base. It is configured to receive multi-dimensional fault feature vectors, and combine them with the original fault codes and their corresponding timestamps to query the fault knowledge base to obtain standard fault identifiers and standard fault mode descriptions. It performs an initial severity assessment based on the multi-dimensional fault feature vectors and encapsulates all information into fault case data units in a predefined structured format.

[0047] The case storage and model iteration module is communication-coupled to the standardized encapsulation module and the fault knowledge base, and includes a fault case library. It is configured to store the received fault case data units into the fault case library, and update the fault diagnosis and prediction model based on the accumulated fault case data units. It is also configured to receive external maintenance feedback information, generate optimization feedback signals by comparing the model diagnosis results with the actual fault causes, and drive the adaptive update of the fault knowledge base based on the optimization feedback signals.

[0048] As can be seen from the technical solutions provided by the present invention above, the centrifugal oil mist lubricator fault code data processing method and system provided by the present invention have the following beneficial effects:

[0049] Precise and efficient data processing: The multi-source data synchronous acquisition mechanism acquires operating status, environmental parameters, and original fault codes in parallel through hardwired connections and sensor networks. Combined with microsecond-level high-precision timestamps and time axis alignment technology, it effectively eliminates abnormal data, ensures data timeliness and consistency, and provides a high-quality data foundation for fault diagnosis. The context-aware association module adaptively defines time windows and key parameters based on the fault knowledge base, accurately extracts fault-related data fragments, avoids interference from irrelevant information, and greatly improves the targeting and efficiency of data processing.

[0050] High accuracy in fault diagnosis: The adaptive feature extraction module extracts features from multiple dimensions, including time domain, frequency domain, and time-frequency domain, based on fault feature templates. This comprehensively captures the essential features of the fault, and the generated multi-dimensional fault feature vector provides a quantitative basis for fault mode recognition. The standardized encapsulation process achieves accurate confirmation of fault modes and scientific assessment of initial severity through fault mode matching and feature threshold comparison. Combined with the joint index structure of the fault case library, it helps to quickly retrieve similar cases, further improving the accuracy and reliability of fault diagnosis.

[0051] The system has strong adaptive evolution capabilities: the case storage and model iteration module triggers incremental model updates based on dynamic changes in the fault case library, continuously optimizes model parameters using new cases, compares and analyzes diagnostic deviations with maintenance feedback information, generates optimization signals to update the fault knowledge base, and realizes adaptive evolution of the model and knowledge base. As cases accumulate, the accuracy of fault diagnosis and prediction is continuously improved, and the stability and adaptability are continuously enhanced in long-term use.

[0052] Convenient and efficient operation and maintenance management: The standardized fault case data unit format is uniform and supports dual retrieval by time and feature dimensions, which facilitates the storage, management and reuse of fault cases. The fully automated data processing from collection, association, extraction, encapsulation to model update reduces manual intervention and lowers operation and maintenance costs. The application of distributed time series database ensures efficient storage and fast access to a large number of fault cases, meets the fault data management needs of long-term equipment operation, and provides comprehensive and powerful support for equipment maintenance decisions. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the steps in the centrifugal oil mist generator fault code data processing method of the present invention;

[0054] Figure 2 This is a schematic diagram of the fault code data processing system for the centrifugal oil mist generator of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0056] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0057] like Figure 1-2 As shown, this embodiment of the invention provides a method for processing fault code data of a centrifugal oil mist lubricator, including the following steps:

[0058] Step S1: Construct and synchronize data acquisition streams: acquire the operating status time-series data stream of the centrifugal oil mist generator, the environmental parameter time-series data stream, and the original fault code sequence triggered by the event in parallel, and uniformly add high-precision timestamps.

[0059] Step S2: Context-aware data association based on fault knowledge base: In response to the capture of an original fault code, a preset fault knowledge base is invoked. Based on the potential fault mode indicated by the fault code, an association time window and a set of key association parameters are adaptively determined. Data fragments of key association parameters within the time window are extracted from the time-series data stream of operating status and environmental parameters to form an original context dataset strongly correlated with the original fault code.

[0060] Step S3, Adaptive feature extraction and vectorization for diagnosis: The original context dataset is preprocessed, and a multidimensional fault feature vector is automatically calculated and generated from the preprocessed data based on the fault feature template defined for the fault mode in the fault knowledge base; the feature vector includes at least the feature quantities extracted from the time domain, frequency domain, or time-frequency domain to distinguish the fault mode;

[0061] Step S4: Generate standardized and scalable fault case units: Encapsulate the original fault code, multi-dimensional fault feature vector, corresponding fault timestamp, standardized fault identifier obtained from the fault knowledge base, and initial severity assessment into a structured fault case data unit.

[0062] Step S5, Fault Case Storage and Model Iterative Update: Store fault case data units in the case library and use them to update or train the fault diagnosis and prediction model; at the same time, based on the difference between the model diagnosis results and subsequent actual maintenance feedback, generate optimization feedback signals for feature templates or association rules in the fault knowledge base to achieve adaptive evolution of the knowledge base.

[0063] In this embodiment, the core function of step S1 is to collect the time-series data streams of the centrifugal oil mist generator's operating status, environmental parameters, and event-triggered original fault code sequences in parallel. Through unified high-precision timestamp allocation, abnormal data removal, and time-axis alignment of multi-source data streams, a standardized synchronous data acquisition stream is generated, providing a high-quality, time-consistent data foundation for subsequent contextual analysis and feature extraction of fault events. The detailed steps are as follows:

[0064] Step S1-1: Acquisition of runtime status timing data stream:

[0065] A hard-wired data acquisition link is established, directly connected to the core operation monitoring node of the centrifugal oil mist generator, and real-time time-series data of four key operating parameters are collected to form an operating status time-series data stream. The collected operating parameters include motor current, motor voltage, fan speed, and inlet / outlet pressure difference. The acquisition process is continuous and uninterrupted, and the acquisition frequency is preset to 100 times per second according to the equipment operating characteristics to ensure that the dynamic changes of parameters are captured. The collected data is transmitted to the data buffer unit in real time in numerical form.

[0066] Step S1-2: Acquisition of environmental parameter time-series data stream:

[0067] A distributed sensor network is deployed to acquire key parameters of the centrifugal oil mist generator's operating environment using an asynchronous acquisition mode, forming a time-series data stream of environmental parameters. The acquired environmental parameters include ambient temperature, ambient humidity, and ambient particulate matter concentration. Sensor nodes send data at a preset cycle, with the acquisition cycle set to 10 times per second, balancing data timeliness and transmission efficiency. The acquired environmental parameter data is aggregated by the sensor network and transmitted to a data cache unit, where it is stored separately from the time-series data stream of the operating status.

[0068] Step S1-3: Original fault code sequence capture:

[0069] The communication bus monitoring process is initiated to continuously monitor the event messages actively reported by the centrifugal oil mist generator controller through the communication bus; the received messages are identified by type, and when the message type is determined to be a fault alarm, the fault code parsing process is triggered to extract the original fault code triggered by the event from the message, and form the original fault code sequence according to the receiving order. At the same time, the message receiving time is recorded to prepare for subsequent timestamp association. The original fault code sequence is written to the independent partition of the data cache unit in real time.

[0070] Step S1-4: Attach high-precision timestamps uniformly:

[0071] The same high-precision hardware clock is introduced as the time reference. The time accuracy of this clock reaches the microsecond level, ensuring the accuracy and consistency of the timestamps. For each data record of the running status time-series data stream, environmental parameter time-series data stream and original fault code sequence stored in the data cache unit, a timestamp generated by this hardware clock is added. The timestamp format is uniformly a continuous digital sequence, which is directly associated with the generation or reception time of the corresponding data record, realizing the unified marking of the time dimension of the three types of data streams.

[0072] Step S1-5: Data caching and outlier removal:

[0073] Temporarily cache the multi-channel data streams with timestamps attached. Configure the cache capacity based on the amount of data generated during one hour of continuous device operation to ensure no data loss. Activate an outlier detection mechanism to verify each data record within a preset reasonable value range. For example, the reasonable range for motor current is set to 0.1 to 2.0 times the device's rated current. Data exceeding this range is judged as obviously invalid outlier data points and directly removed from the data stream, retaining only valid data that conforms to the device's operating patterns.

[0074] Steps S1-6: Time axis alignment and synchronization of data acquisition stream output:

[0075] Based on the unified high-precision timestamps attached to each data record, the timeline alignment of the runtime status time-series data stream and the environmental parameter time-series data stream is performed; let the set of acquisition time points for the runtime status time-series data stream be... The set of collection time points for the environmental parameter time series data stream is as follows: (in, For the running status data The timestamp of the record For environmental parameter data The timestamp of the record For 1-n, For data points of 1-m), linear interpolation is used to supplement missing time points, making the time points of the two types of data streams completely consistent, forming a time-synchronized integrated data stream; the final output is a cleaned synchronous data acquisition stream after outlier removal and time axis alignment, providing a unified data input for subsequent steps.

[0076] In this embodiment, the core function of step S2 is to respond to the capture of the original fault code. By calling a preset fault knowledge base, it adaptively determines the associated time window and key associated parameters that match the fault mode, extracts relevant data fragments within the time window from the synchronous data acquisition stream, and combines and encapsulates them to form an original context dataset strongly correlated with the fault event, providing precise and focused data support for subsequent feature extraction. The detailed steps are as follows:

[0077] Step S2-1: Original fault code capture and fault timestamp recording:

[0078] The original fault code sequence output by the real-time monitoring data synchronization acquisition module is triggered immediately when an original fault code is captured. Based on the high-precision hardware clock used in step S1, the time data of the fault event is extracted, a fault timestamp is generated and recorded in real time. The fault timestamp corresponds one-to-one with the original fault code and serves as the core time reference for subsequent time window calculation and data association.

[0079] Step S2-2: Fault knowledge base query and associated configuration acquisition:

[0080] Using the captured original fault code as an index, a query process is initiated to the pre-set fault knowledge base. The fault knowledge base pre-stores the binding relationships between all known original fault codes and their corresponding fault information. After the query, three types of key information will be obtained: first, a standardized fault identifier that uniquely corresponds to the original fault code, used to achieve unified identification of the fault mode; second, a preset associated time window configuration, including the pre-time length and post-time length of the time window; and third, a predefined list of key associated parameters, which includes the names of operating status parameters and environmental parameters related to the current fault mode. These three types of information are extracted and temporarily stored simultaneously to provide a configuration basis for subsequent operations.

[0081] Step S2-3: Defining the associated time window and calculating the start and end times:

[0082] Based on the fault timestamp recorded in step S2-1 and the associated time window configuration obtained in step S2-2, calculate the start and end times of the associated time window; let the fault timestamp be... The preceding time length in the associated time window configuration is The post-time length is Then the starting time point and termination time point The calculation formula is as follows:

[0083] ;

[0084] ;

[0085] in, The timestamp of the fault event. The length of time before the failure occurs that needs to be included in the context. The length of time after a fault occurs that needs to be included in the context.

[0086] The result obtained by the above formula and The scope of the associated time window was defined, which covers the key data periods related to the fault event before and after the fault occurred;

[0087] Step S2-4: Extraction of time-series data segments for key correlation parameters:

[0088] Load the list of key correlation parameters obtained in step S2-2, and iterate through the synchronous data acquisition stream processed in step S1; the synchronous data acquisition stream has achieved time axis alignment, and is arranged according to the start time point of the correlation time window. and termination time point Filter out the time series data corresponding to all key correlation parameters within the time window; for each key correlation parameter, extract its time series data. to The complete data sequence within a time period is formed into independent time-series data segments; the time order of the data is kept unchanged during the extraction process to ensure that the time-series characteristics of the data are not destroyed.

[0089] Step S2-5: Combining and encapsulating the original context dataset:

[0090] The time-series data fragments of each key correlation parameter extracted in step S2-4 are integrated with the fault timestamp recorded in step S2-1, the original fault code and standardized fault identifier obtained in step S2-2; according to the preset encapsulation format, the various types of data are combined in the order of "original fault code - standardized fault identifier - fault timestamp - set of time-series data fragments of key correlation parameters" to form a structurally unified original context dataset; this dataset only contains data information strongly related to the current fault event, effectively filtering irrelevant data, and providing accurate and focused data input for subsequent feature extraction and fault diagnosis.

[0091] In this embodiment, the core function of step S3 is to perform standardized preprocessing on the original context dataset. Based on the feature templates of the corresponding fault modes in the fault knowledge base, time-domain and frequency-domain features are automatically calculated and extracted. Through ordered combination, multi-dimensional fault feature vectors are generated, transforming unstructured time-series data into structured feature representations, providing quantifiable core evidence for fault mode recognition and severity assessment. The detailed steps are as follows:

[0092] Step S3-1: Preprocessing of the original context dataset:

[0093] Load the original context dataset generated in step S2, start the data preprocessing process, and generate a standardized data sequence through outlier removal noise filtering and data normalization to ensure data quality and consistency.

[0094] Outlier removal employs the 3σ criterion, and statistical analysis is performed on the time-series data of each key correlation parameter to calculate the data mean. and standard deviation It will exceed to Data within the specified range was identified as outliers and removed.

[0095] Noise filtering employs a finite impulse response low-pass filter with a cutoff frequency of 10Hz. This filter is used to process the time series data after outlier removal, filtering out high-frequency interference noise while preserving the trend characteristics of the data.

[0096] Data normalization employs the min-max normalization method, mapping the filtered time-series data to the interval between 0 and 1. The calculation formula is as follows:

[0097] ;

[0098] in, These are the original data values ​​before normalization. This is the minimum value in the time series data for this parameter. This is the maximum value in the time series data for this parameter. These are the normalized standard data values;

[0099] The above preprocessing operations are performed on the time series data of all key correlation parameters one by one, and then integrated to form a standardized data sequence;

[0100] Step S3-2: Fault Feature Template Query:

[0101] Using the standardized fault identifier obtained in step S2 as an index, initiate a query operation on the fault knowledge base; the fault knowledge base pre-stores fault feature templates corresponding to various fault modes, and after querying, obtain the target fault feature template bound to the current standardized fault identifier.

[0102] The fault feature template clearly defines the extraction requirements for three types of feature quantities, including time-domain feature quantity types and calculation methods, frequency-domain feature quantity types and calculation methods, and time-frequency domain feature quantity types and calculation methods.

[0103] Among them, the time-domain features include mean, variance, peak value, peak factor, kurtosis, and skewness; the frequency-domain features include power spectral density, main frequency amplitude, and total harmonic distortion rate; the time-frequency domain features include wavelet decomposition low-frequency coefficient mean, wavelet decomposition high-frequency coefficient variance, and wavelet packet energy entropy. The calculation methods for each type of feature are clearly defined in the template.

[0104] Step S3-3: Multi-domain feature calculation and extraction:

[0105] Based on the definition of the fault feature template, feature calculation operations are performed on each standardized data sequence to extract time-domain features, frequency-domain features, and time-frequency-domain features respectively.

[0106] Calculation of time-domain features:

[0107] The mean is calculated using the arithmetic mean method, and the formula is as follows:

[0108] ,in, To standardize the length of the data sequence, For the first in the standardized data sequence Data points, The mean;

[0109] Variance is calculated using the sample variance formula:

[0110] ,in, This represents the variance; the other symbols have the same meanings as those used in the mean calculation.

[0111] Peak value is the maximum value in the standardized data sequence, peak factor is the ratio of peak value to root mean square value, kurtosis is the ratio of fourth central moment to squared variance, and skewness is the ratio of third central moment to cubed variance. All are calculated according to the standard formula defined in the template to extract the set of time-domain features.

[0112] Frequency domain characteristic calculation:

[0113] First, perform a Fast Fourier Transform on the standardized data sequence to convert the time-domain data into the frequency-domain data, as shown in the following formula:

[0114] ,in, For the first in frequency domain data One frequency component, The imaginary unit, The value range is 1 to ;

[0115] The power spectral density is calculated based on frequency domain data using the periodogram method, as shown in the following formula:

[0116] ,in, For the first Power spectral density of each frequency component for The model;

[0117] Calculation of time-frequency domain features:

[0118] The db4 wavelet is used to perform three-level wavelet decomposition on the standardized data sequence to obtain a low-frequency coefficient sequence and three high-frequency coefficient sequences; the mean of the low-frequency coefficients is the arithmetic mean of the low-frequency coefficient sequence, and the variance of the high-frequency coefficients is the sample variance of the high-frequency coefficient sequence.

[0119] Perform wavelet packet decomposition on the standardized data sequence, setting the decomposition level to 3, and calculate the energy of each wavelet packet subband. Calculate the wavelet packet energy entropy using the following formula:

[0120] ,in, The number of subbands after wavelet packet decomposition. For the first The proportion of energy in the total energy of each individual. The wavelet packet energy entropy;

[0121] The time-frequency domain feature set is extracted using the method described above.

[0122] Step S3-4: Generation of multidimensional fault feature vectors:

[0123] According to the preset feature quantity order in the fault feature template, the time-domain feature quantity, frequency-domain feature quantity and time-frequency-domain feature quantity extracted in step S3-3 are combined in an orderly manner;

[0124] The combination order is as follows: mean, variance, peak value, peak factor, kurtosis, and skewness in the time domain features; power spectral density, main frequency amplitude, and total harmonic distortion rate in the frequency domain features; and wavelet decomposition low-frequency coefficients, mean wavelet decomposition high-frequency coefficients, variance, wavelet packet, and energy entropy in the time and frequency domain features.

[0125] After combination, a fixed-dimensional vector structure is formed, which is the multidimensional fault feature vector. Each element in the vector corresponds to a specific feature value, which is used for subsequent fault mode matching and severity assessment.

[0126] In this embodiment, the core function of step S4 is to complete the matching and confirmation of fault modes and the initial severity assessment based on the association query between the multi-dimensional fault feature vector and the fault knowledge base, integrate all key information related to the fault event, and encapsulate it into standardized and scalable fault case data units according to a predefined structured format, providing a unified and standardized data carrier for fault case storage, retrieval, and subsequent model training; the detailed steps are as follows:

[0127] Step S4-1: Fault Knowledge Base Query and Core Information Acquisition:

[0128] Using the standardized fault identifiers obtained in step S2 as indexes, a precise query process is initiated on the pre-built fault knowledge base. The fault knowledge base pre-stores the mapping relationship between each standardized fault identifier and its corresponding standard fault mode description and formatted storage template. After querying, two types of key information are extracted: one is the standard fault mode description, which clearly defines the occurrence mechanism, typical characteristics, and scope of impact of the corresponding fault, providing a basis for fault event pattern matching; the other is the formatted storage template, which defines the field structure, field type, and field order of fault case data units, ensuring the uniformity of the format of all fault case data units and supporting subsequent standardized storage and retrieval. Both types of information are temporarily stored in the cache unit to provide basic support for subsequent operations.

[0129] Step S4-2: Fault Mode Matching Confirmation and Initial Severity Assessment:

[0130] First, a fault mode matching operation is performed. The core information of the fault event indicated by the original fault code captured in step S1 is loaded and compared with the standard fault mode description obtained in step S4-1 dimension by dimension. The comparison dimensions include fault triggering conditions, trend of related parameter changes, and potentially affected components. If the matching degree of all dimensions reaches the preset threshold (default is set to 95%), the fault mode is confirmed and the standard fault mode to which the fault event belongs is determined. If the matching degree does not reach the threshold, it is marked as a mode to be verified and further confirmed in conjunction with maintenance feedback.

[0131] Subsequently, an initial severity assessment is calculated, and based on the multidimensional fault feature vector generated in step S3, the specific values ​​of each feature quantity are extracted. Simultaneously, the preset reference values ​​of each characteristic quantity corresponding to the standard fault mode are retrieved from the fault knowledge base. The allowable deviation range; the initial severity assessment uses a characteristic quantity deviation weighted calculation method, as shown in the following formula:

[0132] ,in, This is the initial severity assessment value (ranging from 0 to 100, with higher values ​​indicating lower severity of the fault). For the first The weighting coefficients of each feature (the sum of all feature weighting coefficients is 1, preset according to the influence of each feature on the severity of the fault) The first fault feature vector in the multidimensional fault feature vector The actual values ​​of each characteristic quantity This is the preset reference value corresponding to this feature quantity. The total number of features in a multidimensional fault feature vector;

[0133] The deviation contribution value is calculated for each feature in the multidimensional fault feature vector, and the weighted sum is obtained by weighting the values ​​according to the weight coefficients to obtain the final initial severity assessment value, thus completing the preliminary quantitative assessment of the severity of the fault event.

[0134] Step S4-3: Integration and association of all key information:

[0135] Initiate the information integration process to associate and bind all key information related to the current fault event. The information to be integrated includes seven core categories: the fault timestamp recorded in step S1, the standardized fault identifier obtained in step S2, the confirmed fault mode in step S4-2, the initial severity assessment calculated in step S4-2, the multidimensional fault feature vector generated in step S3, and the start time point of the associated time window defined in step S52. and termination time point ;

[0136] During the integration process, fault timestamps and standardized fault identifiers are used as the core association keys to ensure the unique correspondence between various types of information and fault events; the multidimensional fault feature vectors are arranged according to the feature quantity order preset in step S3 to maintain consistency with the fault feature template; the start and end time markers of the associated time window are set according to... in front, The data is appended sequentially to clarify the time frame of the context data; after integration, a structured information set is formed to ensure that no key information is omitted;

[0137] Step S4-4: Standardize the encapsulation of fault case data units:

[0138] Load the formatted storage template obtained in step S4-1. This template contains a fixed list of fields, which are, in order: fault case identifier field, fault timestamp field, standardized fault identifier field, fault mode field, initial severity assessment field, associated time window start and end field, multidimensional fault feature vector field, and original fault code field.

[0139] The information set integrated in step S4-3 is filled one by one according to the field order of the formatted storage template. Each field strictly follows the field type requirements defined in the template. For example, the fault timestamp field uses a microsecond-level numerical format, the initial severity assessment field uses a numerical format with two decimal places, and the multidimensional fault feature vector field uses an array format. After filling, the information of each field is verified to ensure the integrity of the field content and the correctness of the format. After the verification is passed, it is packaged to form a fault case data unit.

[0140] This fault case data unit is configured to support dual search functions. On the one hand, it supports time-dimensional search based on fault timestamps, which can quickly filter fault cases within a specific time period. On the other hand, it supports feature-dimensional search based on fault characteristics, which can find similar fault cases by matching the range of feature values, laying the foundation for the efficient use of the fault case library in the future.

[0141] In this embodiment, the core function of step S5 is to standardize and store the standardized fault case data units and construct an efficient retrieval structure. Based on the incremental sample updates accumulated in the case library, the fault diagnosis and prediction model is updated. The model's diagnostic deviation is compared and analyzed in conjunction with actual maintenance feedback information. An optimized feedback signal is generated to drive the adaptive evolution of the fault knowledge base, continuously improving the accuracy and reliability of fault diagnosis. The detailed steps are as follows:

[0142] Step S5-1: Construction of the storage and retrieval structure for fault case data units:

[0143] Load the fault case data units generated in step S4 and write them in batches into the fault case library of the distributed time series database. The distributed time series database has high throughput and low latency storage characteristics, which can meet the time series storage requirements of fault case data. Assign a globally unique case identifier to each fault case data unit. The identifier adopts a fixed-length numerical encoding form to ensure the unique identification of each case.

[0144] Based on the fault timestamps and standardized fault identifiers in the fault case data units, a composite index structure is established. The index key of the composite index is composed of the fault timestamps and standardized fault identifiers in sequence. Through this index, all fault cases within a specific time range or under a specific fault mode can be quickly located, improving the case retrieval efficiency. After storage is completed, a storage success confirmation signal is generated and fed back to the standardized encapsulation module.

[0145] Step S5-2: Triggering incremental model updates and preparing the input sample set:

[0146] The system monitors the dynamic changes in the fault case library in real time and sets the trigger conditions as the growth rate of the number of cases in the fault case library reaches a preset threshold or a new entry appears in the case identifier set. The preset threshold is configured by default to 10% of the current total number of cases in the case library. When either trigger condition is met, the incremental update process of the fault diagnosis and prediction model is automatically started.

[0147] The input sample set is extracted from the fault case database. The composition of the sample set is divided into two modes: when the total number of cases in the database is small, all fault case data units are extracted; when the total number of cases in the database is large, recently added fault case data units are extracted, with the recent time range preset to 30 days. The extracted fault case data units are parsed to extract the multi-dimensional fault feature vectors contained therein as input samples for model training. The sample sequence is constructed according to the storage order of the fault cases to ensure the temporal characteristics of the samples.

[0148] Step S5-3: Fault diagnosis and prediction model training and performance verification:

[0149] The constructed input sample set is fed into the fault diagnosis and prediction model, and the gradient descent optimization algorithm is used to train or retrain the model. During the training process, the internal weight parameters and bias terms of the model are updated with the goal of minimizing the deviation between the model's prediction results and the actual fault modes in the fault cases. The number of training iterations is set to 200 rounds, the initial learning rate is set to 0.001, and the learning rate decays to 50% of the current value every 50 rounds.

[0150] After training, the reserved validation case set is loaded. The validation case set consists of fault cases proportionally extracted from the fault case library, with an extraction ratio of 20% of the total number of cases in the library, and no overlap with the input sample set. The multidimensional fault feature vector of the validation case set is input into the updated model to obtain the model's diagnostic results. The diagnostic accuracy of the model for each standardized fault identifier is calculated using the following formula:

[0151] ,in, To improve diagnostic accuracy, This refers to the number of cases where the model's diagnostic results match the actual failure modes of the validation cases. To verify the total number of cases in the case set, a confusion matrix is ​​constructed. The rows of the confusion matrix represent the actual failure mode categories of the verification cases, the columns represent the failure mode categories diagnosed by the model, and the matrix element values ​​represent the number of cases in the corresponding category. The confusion matrix can intuitively reflect the diagnostic bias of the model on various failure modes.

[0152] Step S5-4: Receiving and Associating Repair Feedback Information:

[0153] Establish a maintenance feedback information receiving interface to receive actual maintenance work order feedback information uploaded by external systems for recorded fault cases; the feedback information must include three core contents, namely the actual fault cause confirmed by maintenance personnel, the maintenance measures implemented, and the repair verification results after maintenance; perform format verification on the received feedback information to ensure the integrity and standardization of the information. After the verification is passed, extract the case association identifier from the feedback information. This identifier is consistent with the case identifier of the fault case data unit.

[0154] Using the case identifier as the association key, the corresponding fault case data unit is searched in the fault case database. The feedback information is then associated and bound to the fault case data unit to generate an extended fault case data unit containing the feedback information, which is then updated to the fault case database. After the binding is completed, the fault case data unit is marked as having been associated with feedback, which facilitates subsequent traceability and query.

[0155] Step S5-5: Analysis of discrepancies between diagnostic results and maintenance feedback, and optimization of feedback signal generation:

[0156] The extended fault case data unit is parsed to extract the initial diagnosis result of the fault diagnosis and prediction model for the case, as well as the actual fault cause from the associated actual maintenance feedback information; the initial diagnosis result and the actual fault cause are compared dimension by dimension to determine whether they are consistent; if they are completely consistent, the model diagnosis is determined to be without bias, and there is no need to generate an optimization feedback signal; if there is a difference, the difference analysis process is initiated.

[0157] The difference analysis process compares the fault feature templates, key related parameter lists, and related time window configurations corresponding to the initial diagnostic results with the relevant data features corresponding to the actual fault causes to pinpoint the root causes of the differences. Based on the difference analysis results, optimization feedback signals are generated for the fault knowledge base. The optimization feedback signals include correction suggestion types and specific adjustment content. Correction suggestion types are divided into three categories: fault feature template correction, key related parameter list adjustment, and related time window configuration optimization. Specific adjustment content clearly indicates the fields to be modified, the direction of adjustment, and the suggested values.

[0158] Step S5-6: Adaptive updating and evolution of the fault knowledge base:

[0159] The generated optimization feedback signal is transmitted to the fault knowledge base management module to initiate the knowledge base update process. The update process has two modes: manual review and preset rule approval. The default mode is manual review, in which technical personnel evaluate and confirm the correction suggestions in the optimization feedback signal. When the correction suggestions in the optimization feedback signal meet the preset rationality rules, the preset rule approval mode can be enabled to automatically pass the update. The rationality rules include that the adjusted key related parameters are within the reasonable range of the equipment operating parameters and that the time length of the related time window is within the preset range.

[0160] After review or approval, the corresponding entries in the fault knowledge base are updated according to the specific adjustments made based on the optimization feedback signals. If it is a correction of the fault feature template, the feature type or calculation method corresponding to the standardized fault identifier is modified. If it is an adjustment of the list of key related parameters, parameter entries are added or deleted. If it is an optimization of the related time window configuration, the preceding or following time length is adjusted. After the update is completed, a knowledge base update log is generated, recording the update time, updated entries, and adjusted content, so as to realize the adaptive evolution of the fault knowledge base based on case accumulation and feedback.

[0161] The centrifugal oil mist lubricator fault code data processing system's core function is to synchronously collect fault-related multi-source data, accurately correlate context, adaptively extract multi-domain features, and encapsulate standardized cases through the coordinated operation of five functional modules, according to a pre-defined centrifugal oil mist lubricator fault code data processing method. This includes fault case storage, diagnostic model iteration, and adaptive evolution of the fault knowledge base, ensuring the accuracy, standardization, and efficiency of data processing throughout the entire process. This provides comprehensive technical support for the rapid diagnosis, prediction, and repair of centrifugal oil mist lubricator faults. The specific functions and operation flow of each module are as follows:

[0162] Data synchronization acquisition module:

[0163] The core function of the data synchronization acquisition module is to capture the operating status data, environmental parameter data and original fault codes of the centrifugal oil mist generator in parallel. Through unified timestamp marking and data preprocessing, it outputs a time-aligned and cleaned synchronous data acquisition stream to provide high-quality data input for subsequent modules.

[0164] The specific operation process of this module is as follows:

[0165] Operational status time-series data stream acquisition: A dedicated data acquisition link is built using hard-wired connection, directly connected to the core operation monitoring node of the centrifugal oil mist generator, continuously and uninterruptedly acquiring four key operating parameters: motor current, motor voltage, fan speed, and inlet / outlet pressure difference; the acquisition frequency is preset to 100 times per second to ensure accurate capture of dynamic parameter changes, and the acquired numerical data is transmitted in real time to the module's built-in data cache unit;

[0166] Environmental parameter time-series data stream acquisition: Deploy a distributed sensor network and use an asynchronous acquisition mode to acquire key parameters of the device's operating environment; the acquisition objects include ambient temperature, ambient humidity, and ambient particulate matter concentration. Sensor nodes send data at a preset cycle of 10 times per second, balancing timeliness and transmission efficiency; after the acquired data is aggregated by the sensor network, it is transmitted to the data cache unit and the operating status time-series data stream partition for storage.

[0167] Original fault code sequence capture: Start the communication bus monitoring process to continuously monitor the event messages actively reported by the centrifugal oil mist generator controller through the communication bus; perform type identification on the received messages; when it is determined to be a fault alarm message, immediately trigger the fault code parsing process, extract the original fault codes from the message and form a sequence according to the receiving order, and record the message receiving time, and write the sequence into the independent partition of the data cache unit in real time.

[0168] Unified high-precision timestamp addition: The same high-precision hardware clock is introduced as the time reference. This clock has microsecond-level time accuracy to ensure the accuracy and consistency of time stamps. For each data record of the running status time-series data stream, environmental parameter time-series data stream and original fault code sequence stored in the data cache unit, a unified format digital timestamp generated by this hardware clock is added, which is directly associated with the time of data generation or reception, so as to achieve the unification of the time dimension of the three types of data streams.

[0169] Data preprocessing and synchronous stream output: Multiple data streams with added timestamps are temporarily cached, with the cache capacity configured based on the amount of data generated during one hour of continuous device operation; an outlier detection mechanism is activated, validating each data point within a preset reasonable range of device operating parameters and removing obviously invalid outlier data points; based on a unified timestamp, linear interpolation is used to align the timeline of the running status time-series data stream and the environmental parameter time-series data stream, supplementing missing time point data to ensure complete consistency between the two types of data streams; finally, a cleaned synchronous data acquisition stream, after outlier removal and timeline alignment, is output and synchronously transmitted to the context-aware association module.

[0170] Context-aware association module:

[0171] The core function of the context-aware association module is to communicate and couple the data synchronization acquisition module with the fault knowledge base. In response to the original fault code capture event, it determines the association configuration by querying the knowledge base, extracts relevant data fragments from the synchronization data acquisition stream, and combines them to generate an original context dataset that is strongly correlated with the fault event.

[0172] The specific operation process of this module is as follows:

[0173] Fault timestamp recording: The original fault code sequence output by the real-time monitoring data synchronization acquisition module is used to capture an original fault code. When a fault code is captured, the time data of the fault event is immediately extracted based on the system's unified high-precision hardware clock, a fault timestamp is generated and recorded in real time to ensure the accurate binding of the fault event with the time reference.

[0174] Fault knowledge base query: Using the captured original fault code as an index, a query request is sent to the fault knowledge base; the knowledge base returns three types of key information, namely the standardized fault identifier that uniquely corresponds to the original fault code, the preset association time window configuration, and the predefined list of key association parameters. The three types of information are synchronously stored in the module's temporary cache.

[0175] Defining the associated time window: Based on the fault timestamp of the record and the associated time window configuration obtained from the query, calculate the start time point and the end time point of the associated time window; the start time point is the fault timestamp minus the preceding time length in the configuration, and the end time point is the fault timestamp plus the following time length in the configuration. The range of context data extraction is clearly defined by the start and end time points.

[0176] Key parameter time series data fragment extraction: Load the list of key related parameters obtained by query, traverse the synchronous data acquisition stream output by the data synchronous acquisition module; according to the defined related time window, filter out the complete time series data corresponding to all key related parameters in the window, extract each parameter separately to form an independent time series data fragment, and keep the data time order unchanged during the extraction process;

[0177] The original context dataset is combined and encapsulated: the extracted time-series data fragments of each key correlation parameter are integrated with the recorded fault timestamps, captured original fault codes and queried standardized fault identifiers; various types of data are combined in a fixed order and encapsulated to form a structurally unified original context dataset, ensuring that the dataset contains only information strongly related to the current fault event, and then the dataset is transmitted to the feature adaptive extraction module.

[0178] Feature adaptive extraction module:

[0179] The core function of the feature adaptive extraction module is to communicate and couple the context-aware association module with the fault knowledge base. It receives the original context dataset and performs standardized preprocessing. Based on the fault feature template, it automatically extracts multi-domain features and combines them to generate a multi-dimensional fault feature vector.

[0180] The specific operation process of this module is as follows:

[0181] Original context dataset preprocessing: Load the original context dataset transmitted by the context-aware association module and start the preprocessing process; use the 3σ criterion to remove outliers in the time series data of each key association parameter, filter high-frequency interference noise through a finite impulse response low-pass filter with a cutoff frequency of 10Hz, and then use the min-max normalization method to map the filtered data to the interval of 0 to 1 to eliminate dimensional differences; perform the above operations on the time series data of all key association parameters one by one, and integrate them to form a standardized data sequence;

[0182] Fault Feature Template Query: Using the standardized fault identifier obtained by the context-aware association module as an index, a query is initiated to the fault knowledge base to obtain the fault feature template bound to the standardized fault identifier. The template clearly defines the types of features to be extracted in the time domain, frequency domain, and time-frequency domain, as well as the calculation methods. The time domain features include mean, variance, peak value, peak factor, kurtosis, and skewness. The frequency domain features include power spectral density, dominant frequency amplitude, and total harmonic distortion rate. The time-frequency domain features include the mean of low-frequency coefficients of wavelet decomposition, the variance of high-frequency coefficients of wavelet decomposition, and the energy entropy of wavelet packets.

[0183] Multi-domain feature calculation and extraction: Based on the definition of the fault feature template, feature calculation operations are performed on the standardized data sequence; time-domain features are calculated using the arithmetic mean method, sample variance formula, etc., to obtain the mean, variance, and other indicators. The peak value is the maximum value of the sequence, and the peak factor, kurtosis, and skewness are calculated according to the template's preset standard formula; frequency-domain features are first converted from time-domain data to frequency-domain data through fast Fourier transform, and then the power spectral density is calculated using the periodogram method, while the main frequency amplitude and total harmonic distortion rate are extracted; time-frequency domain features are obtained by using db4 wavelet three-level decomposition to obtain the mean of low-frequency coefficients and the variance of high-frequency coefficients, and the energy proportion of each sub-band is calculated through three-level wavelet packet decomposition to obtain the wavelet packet energy entropy;

[0184] Multidimensional fault feature vector generation: According to the preset feature quantity order in the fault feature template, the extracted time-domain features, frequency-domain features, and time-frequency-domain features are combined in an orderly manner. The combination order is as follows: mean, variance, peak value, peak factor, kurtosis, and skewness of the time-domain features; power spectral density, main frequency amplitude, and total harmonic distortion rate of the frequency-domain features; mean of low-frequency coefficients of wavelet decomposition, variance of high-frequency coefficients of wavelet decomposition, and wavelet packet energy entropy of the time-frequency-domain features. This forms a fixed-dimensional multidimensional fault feature vector, which is then transmitted to the standardized encapsulation module.

[0185] Standardized packaging module:

[0186] The core functions of the standardized encapsulation module are the communication coupling feature adaptive extraction module, the context-aware association module, and the fault knowledge base. It receives multi-dimensional fault feature vectors, integrates all fault-related information, completes fault mode confirmation and initial severity assessment, and encapsulates them into standardized and scalable fault case data units.

[0187] The specific operation process of this module is as follows:

[0188] Standard Fault Information Query: The multi-dimensional fault feature vector transmitted by the response feature adaptive extraction module is used as an index to query the fault knowledge base and obtain the standard fault mode description and formatted storage template corresponding to the standardized fault identifier. The standard fault mode description clarifies the fault occurrence mechanism, typical characteristics and impact range, and the formatted storage template defines the field structure, type and order of the fault case data unit.

[0189] Fault mode matching and initial severity assessment: Based on the standard fault mode description obtained from the query, the core information of the fault event indicated by the original fault code is compared with it in multiple dimensions, including fault triggering conditions, trends of related parameter changes, and potentially affected components. When the matching degree reaches the preset standard, the fault mode is confirmed. At the same time, the specific values ​​of each feature quantity in the multi-dimensional fault feature vector are extracted and compared with the preset threshold of the feature quantity corresponding to the standard fault mode in the fault knowledge base. The initial severity assessment value is generated by weighted calculation of the feature quantity deviation. The assessment value ranges from 0 to 100, with a higher value indicating a lower fault severity.

[0190] Fault-related information integration: The fault timestamp, standardized fault identifier, confirmed fault mode, initial severity assessment, and multidimensional fault feature vector are linked and integrated, and the start and end time points of the associated time window are marked. The integration process uses the fault timestamp and standardized fault identifier as the core association key to ensure that all types of information correspond uniquely to the fault event, and that the multidimensional fault feature vector maintains the same order as the fault feature template.

[0191] Standardized Fault Case Data Unit Encapsulation: Following the formatted storage template obtained from the query, all integrated information, including the original fault codes, is populated into the corresponding fields one by one. Each field strictly adheres to the type requirements defined in the template. The fault timestamp uses a microsecond-level numerical format, the initial severity assessment uses a numerical format with two decimal places, and the multidimensional fault feature vector uses an array format. After population, the field information is checked for completeness and format correctness. Once the check passes, it is encapsulated to form a fault case data unit. This unit supports dual retrieval functions, enabling both time-based retrieval based on the fault timestamp and feature-based retrieval based on fault features. It is then transmitted to the case storage and model iteration module.

[0192] Case storage and model iteration module:

[0193] The core function of the case storage and model iteration module is to communicate with the standardized encapsulation module and the fault knowledge base. It has a built-in fault case library, is responsible for the storage and retrieval structure of fault case data units, drives the incremental update of fault diagnosis and prediction models, receives maintenance feedback and generates optimization signals, and realizes the adaptive evolution of the fault knowledge base.

[0194] The specific operation process of this module is as follows:

[0195] Fault Case Storage and Retrieval Structure Construction: The system receives fault case data units transmitted by the standardized encapsulation module and writes them in batches into the fault case library of the distributed time-series database. Each fault case data unit is assigned a globally unique digital code-based case identifier to ensure unique identification. Based on the fault timestamp and standardized fault identifier in the fault case data unit, a composite index structure is constructed. The index key is formed by combining the two in sequence. This index allows for quick location of fault cases within a specific time range or under a specific fault mode, improving retrieval efficiency.

[0196] Model incremental update triggering and input sample set preparation: Real-time monitoring of the dynamic changes in the fault case library. When the number of cases increases by 10% of the current total or a new entry appears in the case identifier set, the incremental update process of the fault diagnosis and prediction model is automatically triggered. The input sample set is extracted from the fault case library. When the total number of cases in the library is small, all case data units are extracted. When the total number of cases is large, the case data units newly added in the past 30 days are extracted. After parsing, the multi-dimensional fault feature vectors are extracted as input samples for model training. The sample sequence is constructed according to the storage order.

[0197] Fault diagnosis and prediction model training and performance verification: The input sample set is input into the fault diagnosis and prediction model, and gradient descent optimization algorithm is used for training or retraining. The goal is to minimize the deviation between the model prediction results and the actual fault modes of the fault cases. The internal weight parameters and bias terms of the model are updated. The training is set to 200 iterations with an initial learning rate of 0.001, which decays to 50% of the current value every 50 iterations. After training, a reserved validation case set is loaded. This case set is extracted from 20% of the fault case library and has no overlap with the input sample set. The multidimensional fault feature vectors of the validation cases are input into the updated model to obtain the diagnosis results. The diagnostic accuracy corresponding to each standardized fault identifier is calculated. At the same time, a confusion matrix is ​​constructed to intuitively reflect the model's diagnostic bias.

[0198] Maintenance Feedback Reception and Association Binding: An interface for receiving maintenance feedback information is established to receive externally uploaded feedback information from actual maintenance work orders for recorded fault cases. Feedback information must include the confirmed actual fault cause, implemented maintenance measures, and repair verification results. The module performs format validation to ensure completeness and standardization. Case association identifiers are extracted from the feedback information and matched with case identifiers in the fault case database. The feedback information is then associated with the corresponding fault case data unit, generating an extended fault case data unit and updating it to the fault case database. Simultaneously, the case is marked as associated with a feedback status.

[0199] Difference Analysis and Optimization Feedback Signal Generation: The extended fault case data unit is analyzed, and the model's initial diagnostic results for the case are extracted and compared with the associated actual fault causes dimension by dimension. If differences exist, a difference analysis process is initiated. By comparing the fault feature template, key correlation parameter list, and correlation time window configuration corresponding to the initial diagnostic results with the relevant data features corresponding to the actual fault causes, the root cause of the difference is located. Based on the analysis results, optimization feedback signals are generated, specifying the type of correction suggestion, including fault feature template correction, key correlation parameter list adjustment, and correlation time window configuration optimization, while also providing specific adjustment content.

[0200] The fault knowledge base is adaptively updated by transmitting optimization feedback signals to the fault knowledge base management module to initiate the update process. The update process supports two modes: manual review and preset rule approval. The default mode is manual review, and automatic approval can be enabled when the optimization feedback signals meet the preset rationality rules. After the review or approval is passed, the corresponding entries in the fault knowledge base are updated according to the specific content of the optimization feedback signals. This includes modifying the feature type or calculation method of the fault feature template, adjusting the entries in the list of key related parameters, and optimizing the pre- or post-time length of the related time window, so as to realize the adaptive evolution of the fault knowledge base based on case accumulation and feedback.

[0201] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of processing centrifugal oil mist machine fault code data, the method comprising: Includes the following steps: ​ Step S1: Construct and synchronize data acquisition streams: acquire the operating status time-series data stream of the centrifugal oil mist generator, the environmental parameter time-series data stream, and the original fault code sequence triggered by the event in parallel, and uniformly add high-precision timestamps. Step S2, Context-Aware Data Association Based on Fault Knowledge Base: This specifically includes the following steps: In response to the capture of a raw fault code, immediately record the fault timestamp corresponding to the event; Using the original fault code as an index, query the preset fault knowledge base to obtain the standardized fault identifier bound to the fault code, the preset associated time window configuration, and a set of predefined key associated parameters. Based on the fault timestamp and the queried time window configuration, the start and end times for extracting context data are calculated, and the associated time window is defined. Based on the list of key correlation parameters, extract time-series data segments corresponding to all key correlation parameters within the correlation time window from the synchronous data acquisition stream aligned with the time axis; The extracted time-series data fragments of each key correlation parameter are combined and encapsulated with the original fault code, fault timestamp, and standardized fault identifier to generate an original context dataset that is strongly correlated with the original fault code. Step S3, Adaptive feature extraction and vectorization for diagnosis: The original context dataset is preprocessed, and a multidimensional fault feature vector is automatically calculated and generated from the preprocessed data based on the fault feature template defined for the corresponding fault mode in the fault knowledge base; the feature vector includes at least the feature quantity extracted from the time domain, frequency domain or time-frequency domain to distinguish the fault mode; Step S4: Generate standardized and scalable fault case units: Encapsulate the original fault code, multi-dimensional fault feature vector, corresponding fault timestamp, standardized fault identifier obtained from the fault knowledge base, and initial severity assessment into a structured fault case data unit. Step S5, Fault Case Storage and Model Iterative Update: Store fault case data units in the case library and use them to update or train the fault diagnosis and prediction model; at the same time, based on the difference between the model diagnosis results and subsequent actual maintenance feedback, generate optimization feedback signals for feature templates or association rules in the fault knowledge base to achieve adaptive evolution of the knowledge base.

2. The centrifugal oil mist machine fault code data processing method of claim 1, wherein: Step S1, which involves constructing and synchronizing the data acquisition stream, includes: The operating status time-series data stream of the centrifugal oil mist generator is collected in real time through hard-wiring, including motor current, motor voltage, fan speed, and inlet / outlet pressure difference. Meanwhile, the environmental parameters time-series data stream of ambient temperature, ambient humidity, and ambient particulate matter concentration are collected asynchronously through a sensor network. Listen to the event messages actively reported by the centrifugal oil mist generator controller through the communication bus. When the message type is a fault alarm, parse out the original fault code sequence that triggered the event. For each data record of the running status timing data stream, the environmental parameter timing data stream, and the original fault code sequence, a microsecond-level timestamp originating from the same high-precision hardware clock is added; The system caches and preprocesses the timestamped multi-channel data, removes obviously invalid and abnormal data points, and aligns the time-series data streams of operating status and environmental parameters at different acquisition frequencies with the time-series data streams based on the timestamps, outputting a cleaned synchronous data acquisition stream.

3. The centrifugal oil mist machine fault code data processing method of claim 1, wherein: Step S3, the adaptive feature extraction and vectorization for diagnosis, specifically includes the following steps: The original context dataset is preprocessed to generate a preprocessed, standardized data sequence. Data preprocessing includes outlier removal, noise filtering, and data normalization. Using standardized fault identifiers as indexes, the fault knowledge base is queried to obtain the fault feature template corresponding to the fault mode; the fault feature template defines the types of features to be extracted from the time domain, frequency domain, and time-frequency domain, as well as the calculation methods; Based on the fault feature template, feature calculation operations are performed on the preprocessed standardized data sequence to automatically extract time-domain features, frequency-domain features, and time-frequency-domain features. The extracted time-domain features, frequency-domain features, and time-frequency-domain features are combined in a preset order to generate a multidimensional fault feature vector.

4. The centrifugal oil mist machine fault code data processing method of claim 1, wherein: Step S4 specifically includes: In response to the multidimensional fault feature vector, the fault knowledge base is queried using the standardized fault identifier as an index to obtain the standard fault mode description and formatted storage template corresponding to the standardized fault identifier. Based on the standard fault mode description, the fault events indicated by the original fault codes are pattern matched and confirmed, and an initial severity assessment is calculated and generated based on the comparison results of the feature values ​​in the multidimensional fault feature vector with the preset threshold. Associate and integrate fault timestamps, standardized fault identifiers, confirmed fault modes, initial severity assessments, and multidimensional fault feature vectors, and attach start and end time markers for the associated time windows; Based on the formatted storage template, the integrated information, including the original fault codes, is encapsulated into fault case data units with a unified field structure. The fault case data units are configured to support time-based indexing and fault feature-based retrieval.

5. The centrifugal oil mist machine fault code data processing method of claim 1, wherein: Step S5, which involves storing fault cases and iteratively updating the model, specifically includes the following steps: Fault case data units are written into the fault case library of the distributed time series database, and a globally unique case identifier is assigned to each fault case data unit to establish a fast retrieval structure with fault timestamp and standardized fault identifier as a joint index. In response to an increase in the number of cases in the fault case library or a change in the set of case identifiers, the incremental update process of the fault diagnosis and prediction model is triggered. The incremental update process takes all or recently added fault case data units in the case library and their contained multidimensional fault feature vectors as input samples. Based on the input sample set, the fault diagnosis and prediction model is trained or retrained to generate or update the model's internal parameters; the performance of the updated model is verified using the reserved validation case set, and the diagnostic accuracy and confusion matrix of the model for each standardized fault identifier are calculated. Receive feedback information from actual repair work orders for recorded fault cases. The feedback information includes at least the confirmed actual fault cause, repair measures, and repair verification results. Associate and bind the feedback information with the corresponding fault case data unit in the fault case library through the case identifier. The initial diagnostic results of the fault diagnosis and prediction model for a specific fault case are compared with the actual fault causes confirmed in the subsequent actual maintenance feedback information. If there is a difference between the two, the difference analysis process is initiated to generate an optimization feedback signal for the fault knowledge base. The optimization feedback signal includes at least the correction suggestions for the fault feature templates associated with specific standardized fault identifiers, or the adjustment suggestions for the list of key associated parameters and the configuration of associated time windows. Based on the optimization feedback signals, after manual review or approval by preset rules, the corresponding entries in the fault knowledge base are updated, including modifying the fault feature template, adjusting the list of key related parameters, or optimizing the configuration of the related time window, so as to achieve the adaptive evolution of the fault knowledge base based on case accumulation and feedback.

6. A fault code data processing system for centrifugal oil mist lubricator, characterized in that: The system is used to execute the centrifugal oil mist generator fault code data processing method according to any one of claims 1-5, including: The data synchronization acquisition module is configured to acquire the time-series data streams of the centrifugal oil mist generator's operating status and environmental parameters in parallel, and to listen to the event messages reported by the controller to parse the original fault code sequence. It also adds a unified high-precision timestamp to all data, performs time axis alignment and cleaning on multiple data streams, and outputs a synchronous data acquisition stream. The context-aware association module is communication-coupled to the data synchronization acquisition module and a fault knowledge base. It is configured to respond to receiving the original fault code and its corresponding timestamp from the data synchronization acquisition module, query the fault knowledge base to obtain the association time window configuration and key association parameter list, define the time window based on the timestamp and the association time window configuration, and extract the time-series data fragments of key association parameters within the time window from the synchronous data acquisition stream to generate the original context dataset corresponding to the original fault code. The feature adaptive extraction module is communicatively coupled to the context-aware association module and the fault knowledge base. It is configured to receive the original context dataset, preprocess it to generate a standardized data sequence, and automatically calculate and extract time-domain, frequency-domain and time-frequency-domain feature quantities from the standardized data sequence based on the fault feature template obtained from the fault knowledge base that corresponds to the original fault code, and combine them to generate a multi-dimensional fault feature vector. The standardized encapsulation module is communication-coupled to the feature adaptive extraction module, the context-aware association module, and the fault knowledge base. It is configured to receive multi-dimensional fault feature vectors, and combine them with the original fault codes and their corresponding timestamps to query the fault knowledge base to obtain standard fault identifiers and standard fault mode descriptions. It performs an initial severity assessment based on the multi-dimensional fault feature vectors and encapsulates all information into fault case data units in a predefined structured format. The case storage and model iteration module is communication-coupled to the standardized encapsulation module and the fault knowledge base, and includes a fault case library. It is configured to store the received fault case data units into the fault case library, and update the fault diagnosis and prediction model based on the accumulated fault case data units. It is also configured to receive external maintenance feedback information, generate optimization feedback signals by comparing the model diagnosis results with the actual fault causes, and drive the adaptive update of the fault knowledge base based on the optimization feedback signals.

Images