Diesel engine failure early warning method and processing terminal
By acquiring real-time diesel engine data, performing feature extraction and unsupervised early warning model processing, and generating early warning signals and health indices, the problem of real-time early warning and health assessment in diesel engine operation and maintenance is solved, enabling accurate early warning and maintenance suggestions, and improving the operational stability and reliability of diesel engines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGXI FANGCHENGGANG NUCLEAR POWER
- Filing Date
- 2026-05-08
- Publication Date
- 2026-07-31
AI Technical Summary
Existing diesel engine maintenance methods cannot capture changes in equipment status in real time, resulting in frequent false alarms, missed alarms, unplanned downtime, and high maintenance costs. They also cannot issue early warnings in the early stages of a fault, and traditional threshold alarm systems cannot adapt to changes in operating conditions.
A diesel engine fault early warning method is constructed. By acquiring real-time data, performing feature extraction and feature vector data processing, using an unsupervised early warning model to determine the abnormal score set and early warning threshold, generating an early warning signal, and outputting health level and maintenance suggestions based on health index.
It enables accurate early warning of diesel engine faults, outputs health levels and maintenance suggestions, guides operation and maintenance decisions, avoids unplanned downtime, and improves operational stability and reliability.
Smart Images
Figure CN122493633A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of diesel engine technology, and in particular to a diesel engine fault early warning method and processing terminal. Background Technology
[0002] In nuclear power plants, diesel engines are the core power equipment for emergency power supply, and their operational reliability is directly related to production safety and economic benefits. Currently, diesel engine maintenance mainly relies on the following methods: 1. Manual periodic inspection: Maintenance personnel visually observe instruments and record operating parameters, or periodically collect data for offline analysis. This method relies entirely on human experience, cannot capture changes in equipment status in real time, and different personnel have different interpretation standards, which easily leads to missed or false alarms; 2. Post-fault analysis: Historical data is retrieved only after the equipment fails and stops to trace the cause. By this time, the fault has already caused unplanned downtime, production interruption, and high maintenance costs. This is a typical "locking the stable door after the horse has bolted" maintenance method, which cannot issue early warnings in the early stages of a fault and can only respond passively after the fault occurs, resulting in economic losses; 3. Traditional threshold alarm system: Some diesel engines are equipped with simple threshold alarm functions. When parameters exceed fixed limits (such as temperature and pressure), an alarm is triggered. However, fixed thresholds cannot adapt to changes in operating conditions (such as load fluctuations and changes in ambient temperature), resulting in frequent false alarms. Maintenance personnel gradually lose trust in the alarms. This method looks at a single parameter in isolation, making it difficult to comprehensively judge the overall health status of the equipment and provide quantitative and intuitive maintenance suggestions. Therefore, there is an urgent need for a technical solution that can provide early warning of diesel engine failures, evaluate the health status of diesel engines, and provide maintenance suggestions. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a diesel engine fault early warning method and processing terminal.
[0004] The technical solution adopted by this invention to solve its technical problem is: constructing a diesel engine fault early warning method, comprising: Acquire real-time data of the diesel engine; wherein the real-time data includes real-time monitoring data of multiple source signals; Feature extraction is performed on the real-time data to obtain feature vector data; The set of anomaly scores corresponding to each combination of source signals is determined based on the feature vector data; Determine the warning threshold corresponding to each combination of source signals; A corresponding warning signal is generated based on the set of anomaly scores and warning thresholds for each combination of source signals; The health index is determined based on the set of abnormal scores from various combinations of source signals; The health index is used to output the health level and maintenance recommendations.
[0005] Preferably, determining the set of anomaly scores corresponding to various combinations of source signals based on the feature vector data includes: The feature vector data is input into a pre-trained unsupervised early warning model to obtain anomaly score sets corresponding to various combinations of source signals; wherein, the unsupervised early warning model is used to determine the anomaly score sets of various combinations of source signals based on the feature vector data.
[0006] Preferably, the training process of the unsupervised early warning model includes: Acquire training data, which includes sample data of multiple source signal combinations; Feature extraction is performed on the training data to obtain a training set; The feature dimension of each source signal combination is determined based on the number of sensors that collect real-time monitoring data for each source signal combination; An initial model is established based on the isolated forest algorithm and the feature dimensions of each source signal combination; the initial model is configured to determine the distribution of each sample in the sample data and calculate the anomaly score corresponding to each sample based on the distribution of the samples. The initial model is trained using the training set until it meets the set conditions, thus obtaining the unsupervised early warning model.
[0007] Preferably, determining the warning threshold corresponding to each combination of source signals includes: For each source signal combination, the following steps are performed: the abnormal scores corresponding to all samples in the source signal combination are arranged from smallest to largest to obtain an ascending sequence; the index position is determined according to the preset target percentile; and the warning threshold is extracted from the ascending sequence according to the index position.
[0008] Preferably, determining the health index based on the set of abnormal scores from various combinations of source signals includes: For each source signal combination, the abnormal score set is mapped to the percentile of the ascending sequence corresponding to that source signal combination. The average percentile of each abnormal score in each source signal combination is calculated to obtain the health index corresponding to that source signal combination.
[0009] Preferably, generating a corresponding warning signal based on the set of anomaly scores for each combination of source signals and the warning threshold includes: For each source signal combination's set of abnormal scores, the following steps are performed: determine whether each abnormal score in the set is less than the warning threshold corresponding to that source signal combination; if at least one abnormal score is less than the warning threshold corresponding to that source signal combination, generate a warning signal corresponding to that source signal combination.
[0010] Preferably, the step of outputting the health level and maintenance suggestions based on the health index includes: The range of the health index is determined according to a preset level and suggestion list; wherein, the level and suggestion list includes multiple preset levels and maintenance suggestions corresponding to each preset level. Output the health level and maintenance suggestions based on the specified range.
[0011] Preferably, before performing feature extraction on the real-time data, the method further includes: Determine whether a diesel engine start-up identification signal has been obtained; When the diesel engine start-up identification signal is obtained, the start-up time of the diesel engine is determined based on the diesel engine start-up identification signal; The time after the set delay time for the start time is defined as the valid start time; Data from the real-time data that is earlier than the effective start time will be removed.
[0012] Preferably, before performing feature extraction on the real-time data, the method further includes: Acquire the status data of the relevant sensors that collect the real-time data; Determine whether each sensor is in a failed state based on the status data; When there are sensors that are in a malfunctioning state, the data collected by all sensors in a malfunctioning state will be removed from the real-time data.
[0013] Furthermore, the present invention also constructs a processing terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the diesel engine fault early warning method described above.
[0014] Implementing this invention has the following beneficial effects: it can provide accurate early warning of diesel engine failures, and by outputting health levels, it enables maintenance personnel to intuitively and easily understand the health status of the diesel engine, obtain maintenance suggestions, guide maintenance decisions, avoid unplanned downtime, and maximize the value of data. Attached Figure Description
[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a diesel engine fault early warning method in some embodiments of the present invention; Figure 2 This is a schematic diagram of the processing terminal in some embodiments of the present invention. Detailed Implementation
[0016] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0017] It should be noted that the flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0018] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0019] Figure 1 This is a flowchart of a diesel engine fault early warning method in some embodiments of the present invention. This diesel engine fault early warning method can accurately predict diesel engine faults and, by outputting a health level, allows maintenance personnel to intuitively understand the diesel engine's health status, receive maintenance recommendations, guide maintenance decisions, avoid unplanned downtime, and maximize data value. It is of great significance for improving the operational stability and reliability of diesel engines.
[0020] The diesel engine fault early warning method and processing terminal may include steps S10 to S70.
[0021] Step S10 includes: acquiring real-time data of the diesel engine. The real-time data includes real-time monitoring data from a combination of multiple source signals.
[0022] Specifically, the types of source signal combinations may include at least two of the following: lubricating oil pressure group, cylinder exhaust temperature group, turbocharger temperature group, and splash oil temperature group. Specifically, the real-time monitoring data for the lubricating oil pressure group may include detection data output from corresponding pressure sensors that detect the lubricating oil pressure of the first important structural component in the diesel engine (which may include main bearings, camshaft bearings, and piston cooling nozzles, etc.); the real-time monitoring data for the cylinder exhaust temperature group may include detection data output from corresponding temperature sensors that detect the exhaust temperatures of the cylinder in rows A and B; the real-time monitoring data for the turbocharger temperature group may include detection data output from corresponding temperature sensors that detect the inlet and outlet temperatures of the turbocharger in rows A and B, respectively; and the real-time monitoring data for the splash oil temperature group may include detection data output from corresponding temperature sensors that detect the temperature of lubricating oil splashed from the second important structural component (which may include the crankcase, connecting rod big end, and camshaft, etc.).
[0023] Understandably, this step, by acquiring output data from multiple sensors, provides more comprehensive reference data for subsequent steps to more comprehensively realize diesel engine fault early warning and judge the overall health status of the diesel engine, thereby improving the reliability of the early warning results and health evaluation results of the present invention.
[0024] Step S20 includes: extracting features from real-time data to obtain feature vector data.
[0025] In some embodiments, feature extraction of real-time data may include: extracting predetermined features from the output data of each sensor in the real-time data. The predetermined features include at least one of first-order difference values, sliding window mean, and sliding window standard deviation.
[0026] The expression for the first-order difference value can be: , This represents the first-order difference value of the i-th sensor at timestamp t (which reflects the rate of change of the signal value). This represents the signal value output by the i-th sensor at timestamp t. This represents the detection data output by the i-th sensor at timestamp t-1.
[0027] The expression for the sliding window mean can be: , Indicates window width The mean value of the sliding window of the i-th sensor at timestamp t (which reflects the recent trend of the signal value). Indicates the window width. This represents the signal value output by the i-th sensor at timestamp tj.
[0028] The expression for the standard deviation of the sliding window can be: , Indicates window width The standard deviation of the sliding window of the i-th sensor at timestamp t (which reflects the fluctuation of the signal value).
[0029] Step S30 includes: determining the set of anomaly scores corresponding to each combination of source signals based on the feature vector data.
[0030] In some embodiments, the anomaly score sets corresponding to various combinations of source signals can be determined by inputting feature vector data into a pre-trained unsupervised early warning model to obtain the anomaly score sets corresponding to various combinations of source signals. The unsupervised early warning model is used to determine the anomaly score sets for various combinations of source signals based on the feature vector data.
[0031] In this embodiment, the unsupervised early warning model can accurately determine the set of abnormal scores without training with fault samples, and it also features high efficiency and accurate output. It should be noted that the probability of nuclear power plant accidents is low in reality, so fault samples are scarce in industrial scenarios. Theoretically, although simulated fault samples can be obtained through simulators, simulators have the risks of high acquisition costs and discrepancies with real operating conditions.
[0032] In some embodiments, the unsupervised early warning model can be obtained by performing steps S301 to S305.
[0033] Step S301 includes: acquiring training data, which includes sample data of multiple source signal combinations.
[0034] In this step, the sample data can be historical monitoring data from previous operation of the diesel engine, showing the detection of various source signal combinations by relevant sensors. Each sample data corresponds to one of the source signal combinations, including the detection data output by the relevant sensor. The detection data output by each sensor can be considered as a sample.
[0035] Step S302 includes: extracting features from the training data to obtain a training set. In this step, the process of extracting features from the training data can be referred to above, ensuring that the type of the feature vector obtained after feature extraction is the same as the type of the feature vector data.
[0036] Step S303 includes: determining the feature dimension of each source signal combination based on the number of sensors that collect real-time monitoring data for each source signal combination.
[0037] In this embodiment, the feature dimension can be four times the number of sensors. For example, in some embodiments, the acquisition of real-time monitoring data for the lubricating oil pressure group involves three sensors, so the feature dimension for the lubricating oil pressure group can be 12. Furthermore, the acquisition of real-time monitoring data for the cylinder exhaust temperature group involves 18 sensors, the acquisition of real-time monitoring data for the turbocharger temperature group involves four sensors, and the acquisition of real-time monitoring data for the splash oil temperature group involves nine sensors. It should be noted that setting the feature dimension to four times the number of sensors has been verified. This allows for efficient determination of the anomaly score set while ensuring high accuracy of the unsupervised early warning model and reducing the model's training cost.
[0038] Step S304 includes: establishing an initial model based on the isolated forest algorithm and the feature dimensions of each source signal combination; the initial model is configured to determine the distribution of each sample in the sample data, and calculate the anomaly score corresponding to each sample based on the distribution of the samples.
[0039] In this step, the core idea of the initial model is to construct multiple random binary trees using the Isolation Forest algorithm, calculate the path length (the number of edges from the root node to a leaf node) in each tree, then average the path lengths of all trees, and normalize this average with the total number of samples to obtain anomaly scores. The initial model can separate outliers from the training set with a relatively small number of random iterations (i.e., efficiently isolate outliers).
[0040] Furthermore, the expression for the abnormality score can be: Where N represents the number of samples in the training set. This represents the anomaly score corresponding to sample x. Indicates the average path length. Let represent the path length in the t-th tree among the T (natural numbers greater than 1) random binary trees constructed by the Isolation Forest algorithm. Represents the normalization factor. Represents the harmonic number. , Understandably, in calculation Let i = N-1. The set of outlier scores contains the outlier scores of the relevant samples. It is easy to understand that the closer the outlier score is to 1, the easier it is for the sample to be isolated, i.e., an outlier; the closer the outlier score is to 0, the harder it is for the sample to be isolated, i.e., a normal sample.
[0041] To conform to the engineering practice of "the lower the score, the more abnormal," abnormal scores can be processed in reverse: , This represents the anomaly score after reverse processing. Thus, the final output or displayed anomaly score ranges from [0,1), with lower scores indicating a higher degree of anomaly.
[0042] Step S305 includes: training the initial model using the training set until it meets the set conditions to obtain an unsupervised early warning model.
[0043] In this step, meeting the set conditions can be either reaching the height limit or the samples becoming indivisible. Reaching the height limit means that the path length from the root node to the current node in the random binary tree reaches a set threshold. The samples becoming indivisible means that it is impossible to effectively divide the training set by randomly selecting features and split points.
[0044] It should be noted that building an unsupervised early warning model using the Isolation Forest algorithm does not require collecting fault samples for training, exhibits strong generalization ability, and can be achieved by sampling historical monitoring data from sensor outputs. This not only ensures the accuracy of the model output but also helps reduce the cost of model construction. The Isolation Forest algorithm is a mature technology, and the specific training process for unsupervised early warning models can be found in existing techniques, which will not be elaborated here.
[0045] Step S40 includes: determining the warning threshold corresponding to each combination of source signals.
[0046] In some embodiments, the warning threshold corresponding to each source signal combination can be determined by: for each source signal combination, the abnormal scores corresponding to all samples in the source signal combination are arranged from smallest to largest to obtain an ascending sequence; the index position is determined according to the preset target percentile; and the warning threshold is extracted from the ascending sequence according to the index position.
[0047] In this embodiment, the ascending sequence can be represented as: S = {score1, score2, score3, ..., score...} N},score N This represents the outlier score corresponding to the Nth sample, where score1 ≤ score2 ≤ score3, ... ≤ score N The expression for the index position can be: (rounded up), Indicates the index position. This indicates the target percentile (which can be customized as needed, such as setting it to 0.5). "Round up" means that when k is a real number with a decimal point, the value is rounded up. When the value is 8.5, k is 9. After determining the index position, the kth (after reverse processing) abnormal score in the ascending sequence is set as the warning threshold.
[0048] Alternatively, warning thresholds for each source signal combination can be set using a predictive method. For example, the warning threshold for the lubricating oil pressure group can be set to 0.034, the warning threshold for the cylinder exhaust temperature group to 0.04, the warning threshold for the turbocharger temperature group to 0.045, and the warning threshold for the splash oil temperature group to 0.064.
[0049] Step S50 includes: generating a corresponding warning signal based on the set of abnormal scores for each combination of source signals and the warning threshold.
[0050] In some embodiments, a corresponding warning signal can be generated based on the set of abnormal scores and the warning threshold for each source signal combination in the following manner: For each set of abnormal scores of the source signal combination, determine whether each abnormal score in the set of abnormal scores is less than the warning threshold corresponding to the source signal combination; if at least one abnormal score is less than the warning threshold corresponding to the source signal combination, generate a warning signal corresponding to the source signal combination.
[0051] In this embodiment, the warning signal may include the location information of the sensor corresponding to the abnormal score that is less than the warning threshold, so as to assist the operation and maintenance personnel in locating the sensor that may be abnormal and to facilitate the arrangement of relevant handling measures (including sensor signal shielding, sensor repair, etc.).
[0052] Step S60 includes: determining a health index based on the set of abnormal scores of various source signal combinations.
[0053] In some embodiments, based on step S60, the health index can be determined by performing steps S601 and S602.
[0054] Step S601 includes: for each source signal combination's set of abnormal scores, mapping each abnormal score in the abnormal score set to the percentile of the ascending sequence corresponding to that source signal combination.
[0055] Specifically, taking a specific outlier (denoted as score B) in the set of outlier scores of a certain source signal combination (denoted as combination A) as an example, the process of determining the percentile of score B may include: calculating the mean (i.e., the average value of all outlier scores contained in the ascending sequence) and standard deviation (i.e., the standard deviation of all outlier scores contained in the ascending sequence) of the ascending sequence corresponding to combination A; and calculating the percentile of score B based on a predetermined mapping formula, wherein the predetermined mapping formula can be expressed as: , Indicates percentiles, Using the cumulative distribution function of the standard normal distribution, it is easy to understand how to calculate... season . This represents an abnormal score (i.e., score B) determined based on real-time data. This represents the mean of an ascending sequence (i.e., the mean of the ascending sequence corresponding to combination A). It represents the standard deviation of an ascending sequence (i.e., the standard deviation of the outliers contained in the ascending sequence).
[0056] Intuitively, percentiles directly reflect the position of the corresponding sensor state within the historical normal distribution (for example, a 50% percentile indicates a position comparable to the median level of historical normal states). Since the normal distribution of sensor states generally conforms to the central limit theorem, percentiles offer sufficient engineering accuracy when the sample size is adequate, ensuring a high degree of agreement between the actual false alarm rate and the theoretical value. This overcomes the problem of uncontrolled false alarm rates caused by non-normal data distribution in traditional methods.
[0057] Step S602 includes: calculating the average of the percentiles corresponding to each abnormal score in each source signal combination to obtain the health index corresponding to that source signal combination. In this step, calculating the "average of the percentiles corresponding to each abnormal score in each source signal combination" is equivalent to calculating the average of all abnormal scores corresponding to the sensors involved in the real-time data. This average value serves as the health index of the diesel engine, which can intuitively and accurately represent the health level of the diesel engine.
[0058] Step S70 includes: outputting the health level and maintenance recommendations based on the health index.
[0059] In some embodiments, the health level and maintenance recommendations can be output by performing the following steps: determining the range of the health index based on a preset list of levels and recommendations; wherein the list of levels and recommendations includes multiple preset levels and maintenance recommendations corresponding to each preset level; and outputting the health level and maintenance recommendations based on the range.
[0060] Specifically, the intervals can include: Interval 1 [0%, 20%), Interval 2 [20%, 50%), Interval 3 [50%, 70%), Interval 4 [70%, 90%), and Interval 5 [90%, 100%). The preset levels include Excellent, Good, Average, Poor, and Inferior. Maintenance suggestions corresponding to "Excellent" can include "Excellent condition, no abnormalities." For "Good," "Normal condition, continuous monitoring." For "Average," "Average condition, no risk of failure, careful observation." For "Poor," "Significantly deviating from normal, planned maintenance." For "Inferior," "Severely abnormal condition, immediate shutdown and troubleshooting." The first to fifth intervals correspond to "Inferior," "Poor," "Average," "Good," and "Excellent," respectively. For example, a health index of 30% falls within the second interval, therefore the health level is "Poor," and the maintenance suggestion is "Significantly deviating from normal, planned maintenance."
[0061] In this embodiment, by converting abnormal scores into a health index and a five-level health rating, maintenance personnel can quickly determine the equipment status without requiring specialized knowledge. Furthermore, the health rating and maintenance recommendations can be output to a display screen for intuitive presentation.
[0062] Since the diesel engine may experience state fluctuations during the start-up phase, these invalid data may reduce the accuracy of the early warning of the present invention. In order to avoid the above situation, in some embodiments, before performing feature extraction on the real-time data, steps S11 to S14 can be performed to remove invalid data.
[0063] Step S11 includes: determining whether a diesel engine start-up identification signal has been obtained. In this step, the diesel engine start-up identification signal contains information related to the start-up time of the diesel engine.
[0064] Step S12 includes: when a diesel engine start-up identification signal is obtained, determining the start-up time of the diesel engine based on the diesel engine start-up identification signal.
[0065] Step S13 includes: setting the time after the set delay time for the start time as the valid start time. The set time can be set from 10 seconds to 25 seconds, preferably 15 seconds.
[0066] Step S14 includes: removing data from the real-time data that is earlier than the effective start time.
[0067] Understandably, this embodiment can eliminate the impact of state fluctuations during diesel engine startup. Furthermore, if no diesel engine startup identification signal is obtained, all current real-time data is used as the valid segment.
[0068] Since some sensors may malfunction during diesel engine operation (potentially outputting abnormal monitoring data), maintenance personnel will set the status of the malfunctioning sensors to an invalid state upon discovery (this can be achieved by operating the human-machine interface device or by obtaining the status feedback signal output by the sensor). In order to avoid the output signal of the invalid sensor affecting the early warning accuracy of the present invention, in some embodiments, steps S15 to S17 may be included before feature extraction of real-time data.
[0069] Step S15 includes: acquiring the status data of the relevant sensors used to collect real-time data. In this step, the acquired sensor objects include all sensors involved in the real-time monitoring data of each source signal combination, including sensors that are in a failed state. The status data includes the status of all sensors, indicating whether they are in a normal or failed state.
[0070] Step S16 includes: determining whether each sensor is in a failed state based on the status data.
[0071] Step S17 includes: when there is a sensor in a failed state, removing the data collected by all sensors in a failed state from the real-time data.
[0072] Understandably, this embodiment can eliminate the impact of sensors in a failed state. Furthermore, when all sensors related to a certain source signal combination fail, in subsequent execution programs, the execution steps such as feature extraction, anomaly score calculation, and early warning threshold determination of the real-time monitoring data corresponding to that source signal combination can be skipped, thereby not affecting the evaluation of other source signal combinations.
[0073] Figure 2 This is a schematic diagram of the structure of a processing terminal in some embodiments of the present invention. The processing terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the diesel engine fault early warning method provided in the embodiments of the present invention.
[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0075] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0076] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0077] It is understood that the above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can freely combine the above technical features without departing from the concept of the present invention, and can also make several modifications and improvements, all of which fall within the protection scope of the present invention. Therefore, all equivalent transformations and modifications made with respect to the scope of the claims of the present invention should fall within the scope of the claims of the present invention.
Claims
1. A diesel engine fault early warning method, characterized in that, include: Acquire real-time data of the diesel engine; wherein the real-time data includes real-time monitoring data of multiple source signals; Feature extraction is performed on the real-time data to obtain feature vector data; The set of anomaly scores corresponding to each combination of source signals is determined based on the feature vector data; Determine the warning threshold corresponding to each combination of source signals; A corresponding warning signal is generated based on the set of anomaly scores and warning thresholds for each combination of source signals; The health index is determined based on the set of abnormal scores from various combinations of source signals; The health index is used to output the health level and maintenance recommendations.
2. The diesel engine fault early warning method according to claim 1, characterized in that, The step of determining the set of anomaly scores corresponding to various combinations of source signals based on the feature vector data includes: The feature vector data is input into a pre-trained unsupervised early warning model to obtain anomaly score sets corresponding to various combinations of source signals; wherein, the unsupervised early warning model is used to determine the anomaly score sets of various combinations of source signals based on the feature vector data.
3. The diesel engine fault early warning method according to claim 2, characterized in that, The training process of the unsupervised early warning model includes: Acquire training data, which includes sample data of multiple source signal combinations; Feature extraction is performed on the training data to obtain a training set; The feature dimension of each source signal combination is determined based on the number of sensors that collect real-time monitoring data for each source signal combination; An initial model is established based on the isolated forest algorithm and the feature dimensions of each source signal combination; the initial model is configured to determine the distribution of each sample in the sample data and calculate the anomaly score corresponding to each sample based on the distribution of the samples. The initial model is trained using the training set until it meets the set conditions, thus obtaining the unsupervised early warning model.
4. The diesel engine fault early warning method according to claim 3, characterized in that, Determining the warning threshold corresponding to each combination of source signals includes: For each source signal combination, the following steps are performed: the abnormal scores corresponding to all samples in the source signal combination are arranged from smallest to largest to obtain an ascending sequence; the index position is determined according to the preset target percentile; and the warning threshold is extracted from the ascending sequence according to the index position.
5. The diesel engine fault early warning method according to claim 4, characterized in that, The determination of the health index based on the set of abnormal scores from various combinations of source signals includes: For each source signal combination, the abnormal score set is mapped to the percentile of the ascending sequence corresponding to that source signal combination. The average percentile of each abnormal score in each source signal combination is calculated to obtain the health index corresponding to that source signal combination.
6. The diesel engine fault early warning method according to claim 1, characterized in that, The generation of a corresponding early warning signal based on the set of anomaly scores and the early warning threshold for each combination of source signals includes: For each source signal combination's set of abnormal scores, the following steps are performed: determine whether each abnormal score in the set is less than the warning threshold corresponding to that source signal combination; if at least one abnormal score is less than the warning threshold corresponding to that source signal combination, generate a warning signal corresponding to that source signal combination.
7. The diesel engine fault early warning method according to claim 1, characterized in that, The step of outputting a health level and maintenance recommendations based on the health index includes: The range of the health index is determined according to a preset level and suggestion list; wherein, the level and suggestion list includes multiple preset levels and maintenance suggestions corresponding to each preset level. Output the health level and maintenance suggestions based on the specified range.
8. The diesel engine fault early warning method according to any one of claims 1 to 7, characterized in that, Before performing feature extraction on the real-time data, the following steps are also included: Determine whether a diesel engine start-up identification signal has been obtained; When the diesel engine start-up identification signal is obtained, the start-up time of the diesel engine is determined based on the diesel engine start-up identification signal; The time after the set delay time for the start time is defined as the valid start time; Data from the real-time data that is earlier than the effective start time will be removed.
9. The diesel engine fault early warning method according to any one of claims 1 to 7, characterized in that, Before performing feature extraction on the real-time data, the following steps are also included: Acquire the status data of the relevant sensors that collect the real-time data; Determine whether each sensor is in a failed state based on the status data; When there are sensors that are in a malfunctioning state, the data collected by all sensors in a malfunctioning state will be removed from the real-time data.
10. A processing terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the diesel engine fault early warning method as described in any one of claims 1 to 9.