Marine engine abnormality prediction system

Abstract

The marine engine abnormality detection device stores standard waveforms representing the typical in-cylinder pressure waveform of a marine engine for each load factor. Using the measured waveform representing the in-cylinder pressure waveform of the marine engine and the standard waveform corresponding to the load factor of the marine engine, it calculates the integrated value of the absolute difference between the measured waveform and the standard waveform. This integrated value is compared with a threshold value to detect signs of abnormality in the marine engine. [Effect] By comparing the in-cylinder pressure waveform with a standard waveform, the condition of the marine engine can be accurately evaluated.

Description

[Technical Field]

[0001] The present invention relates to a system for detecting abnormal precursors of marine engines, and covers Otto engines in addition to diesel engines, regardless of whether they are four-stroke or two-stroke. [Background Art]

[0002] The applicant has proposed a remote diagnostic system for marine engines (Patent Document 1: Japanese Patent No. 3696326). The system monitors the engine state via a plurality of sensors and transmits sensor signals to a diagnostic computer on land. The diagnostic computer diagnoses the engine state, and notifies the diagnostic result directly to the shipping company that operates the ship or to the ship. This allows the diagnostic computer to replace part of the engine monitoring work performed by engine crew. In addition, signs of engine abnormality can be detected in advance and maintenance can be performed.

[0003] In order to improve diagnostic quality, it is necessary to satisfy two contradictory requirements: reducing false alarms and detecting precursors that may develop into engine abnormalities at an early stage. Regarding detection of abnormal precursors, in the case of marine engines, abnormalities are usually detected and eliminated at an early stage, so the problem is that there is little data on abnormal conditions. For this reason, it is difficult to recognize the boundary between abnormal conditions and normal conditions using, for example, a random forest algorithm. Therefore, shifts from normal-time signals (statistically average signals) are detected based on the distribution of normal-time sensor signals, such as detecting a deviation greater than or equal to Kσ (where K is a constant such as 3, 5, or 6, and σ is the standard deviation). Under these constraints, it is difficult to reduce false alarms and detect abnormal precursors at an early stage.

[0004] Other related prior patents will be described. Patent Document 2 (Japanese Patent No. 7326174) detects oil leakage from an exhaust valve drive mechanism of a marine engine. To this end, it proposes measuring the in-cylinder pressure waveform of each cylinder in association with the crank angle. [Prior Art Documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 3696326 [Patent Document 2] Patent No. 7326174 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] The objective of this invention is to provide a system capable of detecting signs of malfunction in a marine engine. [Means for solving the problem]

[0007] The marine engine abnormality detection device of this invention includes a computer for detecting abnormality signs and receives data related to the marine engine from a ship or the like. The computer stores standard waveforms that represent the typical in-cylinder pressure waveform of the marine engine for each load factor of the marine engine. Here, the standard waveform is, for example, a standard waveform for each individual cylinder. Then, using the measured waveform that represents the in-cylinder pressure waveform of the marine engine and the standard waveforms stored according to the load factor of the marine engine, data is obtained that represents the integrated absolute value of the difference between the measured waveform and the standard waveform for at least one cycle of the marine engine's operation. Then, the obtained data is compared with a threshold to detect an abnormality in the marine engine.

[0008] In processing the in-cylinder pressure waveform, the absolute value of the waveform difference for each crank angle may be integrated as the square norm, first power norm, maximum value norm, etc. Alternatively, the difference between the Fourier transforms of the measured waveform and the standard waveform, or the Fourier transform of the waveform difference, may be calculated. Furthermore, the in-cylinder pressure waveform may be treated as vector data for each crank angle, and the dot product of the measured waveform and the standard waveform may be divided by the norm of the measured waveform and the norm of the standard waveform. The calculated integrated value may be processed as a single data point or as a set of integrated values ​​for each crank angle.

[0009] The in-cylinder pressure waveform is fundamental data representing the state of each cylinder in a marine engine. By comparing this with a standard waveform, the state of the marine engine can be accurately evaluated. Note that the computer used in this specification does not need to be dedicated to the detection system; a cloud-based computer may also be used.

[0010] Preferably, the absolute values ​​of the differences in the in-cylinder pressure waveforms for each crank angle are integrated to obtain the aforementioned data. In this way, only the absolute values ​​of the differences need to be integrated, making the calculation simpler compared to Fourier transform or dot product operations.

[0011] Preferably, standard waveforms for each engine load factor are stored, and the difference between the measured waveform and the standard waveform at approximately the same load factor is accumulated. This prevents the waveform difference due to load factor from affecting the detection of abnormal signs. When the load factor is low, the engine state is often not steady due to port entry and departure, so preferably, the standard waveform and the measured waveform are compared at load factors above a predetermined value.

[0012] Preferably, the distribution of past data for multiple types of data other than the in-cylinder pressure waveform related to the marine engine is also stored. The computer generates thresholds for each of the multiple types of data from the distribution of past data, compares the thresholds with the current data, and detects signs of abnormality in the marine engine or its peripheral equipment according to the type of data that exceeds the threshold.

[0013] By defining thresholds for multiple types of data, it is possible to detect anomalies in multiple maintenance items.

[0014] More preferably, the computer also stores the distribution of past data for a given combination of data from multiple types of data. Then, using the distribution of past data as a reference, it performs relationship analysis on the current data, that is, it evaluates the degree to which the current data deviates from the distribution of past data, thereby detecting signs of malfunction in the marine engine or its peripheral equipment.

[0015] Even when individual sensor signals appear to fluctuate randomly, examining combinations of multiple sensor signals can sometimes reveal early signs of anomalies. Therefore, relationship analysis can detect early signs that would not be apparent from individual sensor signals alone.

[0016] Preferably, the computer stores data from past instances where abnormal signs were detected and the inspection results of the marine engine at that time. For similar abnormal signs, it refers to the data from past instances where abnormal signs were detected and the inspection results at that time to generate suggestions regarding the inspection of the marine engine or its peripheral equipment.

[0017] By referring to actual maintenance results (inspection results on ships) for similar data from the past, appropriate maintenance suggestions can be made. It is unclear whether the detected signs of an anomaly are actually signs of an anomaly. Therefore, by reflecting actual maintenance results in the suggestions, accurate suggestions can be made.

[0018] More preferably, the computer refers to past inspection results regarding similar anomaly warnings, comparing those indicating maintenance is necessary and those indicating no maintenance is necessary. If the results indicating no maintenance is necessary are predominant, the computer reduces the suggested need for maintenance. For example, this can reduce false alarms in cases where an anomaly warning is issued, but an inspection reveals no problems.

[0019] More preferably, the computer stores thresholds for detecting abnormalities based on data showing the in-cylinder pressure waveform, multiple types of data, and predetermined data combinations. Based on the result that maintenance is required during inspection when an abnormality is detected, the threshold is lowered. In this way, it is possible to detect abnormalities that require maintenance even when the change in data from normal values ​​is small. [Brief explanation of the drawing]

[0020] [Figure 1] Schematic diagram of the abnormality prediction system in the example. [Figure 2]Schematic diagram showing main measurement items during engine operation [Figure 3] Schematic diagram showing main measurement items when the engine is stopped [Figure 4] Diagram showing an example of an in-cylinder pressure waveform for a marine engine [Figure 5] Block diagram of a signal processing apparatus for engine diagnosis in an embodiment [Figure 6] Diagram showing analysis of an in-cylinder pressure waveform of an engine in an embodiment [Figure 7] Diagram showing analysis of an in-cylinder pressure waveform of an engine in a modification [Figure 8] Schematic diagram of vector data used in an embodiment [Figure 9] Diagram showing analysis for sensors other than an in-cylinder pressure sensor [Figure 10] Flowchart showing an analysis algorithm based on relationship analysis between a plurality of sensor signals MODE FOR CARRYING OUT THE INVENTION

[0021] An optimal embodiment for carrying out the present invention is shown below. Embodiment

[0022] Figures 1 to 10 show an embodiment of the marine engine abnormality detection system 10. Figure 1 shows the operating environment of the abnormality detection system 10. The ship 2 is equipped with a 4-stroke or 2-stroke marine diesel engine 3 (main engine). In this embodiment, it is a coastal vessel, but it may also be an ocean-going vessel. In this embodiment, the engine has 6 cylinders, and the type of fuel is not limited to heavy oil, but may also be ammonia, hydrogen, methanol, etc. The state of the engine 3 is monitored by multiple sensors, and these multiple sensors are collectively called the sensor set 4. The signals from the sensor set 4 are processed by a computer 5. The sensor set 4 consists of an in-cylinder pressure sensor that measures the in-cylinder pressure waveform for each cylinder of the engine, and multiple other sensors. The signals from the sensor set 4 are displayed on the ship 2's monitor 6. Preferably, in addition to this, reports from the land-based abnormality detection system 10 or instructions from the shipping company's land-based operation center 80 are also displayed. Manual input by the engine operator is also performed on the monitor 6. Signals from the sensors 4 are transmitted via the communication unit 7 to the land-based anomaly detection system 10 at predetermined intervals, such as every 4 hours or every minute. At this time, the current position of the ship 2, destination, weather, etc., are also transmitted. The transmission interval is arbitrary and may be changed according to the status of the engine 3.

[0023] The communication unit 12 of the anomaly prediction detection system 10 communicates with the ship 2 via satellite communication, mobile phone lines, etc., and receives the ship 2's position, weather, sensor signals 4, etc. The system 10 also receives the maintenance results (inspection results) of the engine 3 on the ship 2. The memory unit 14 stores the data acquired from the ship 2 and the results obtained from processing within the system 10. It also stores initial data about the engine 3, such as data from when the engine 3 was tested on land. This data is stored, for example, as vector data, so that it can be easily processed by the AI. The AI ​​may be installed in the computer 5, or the computer 5 may connect to the AI ​​in the cloud.

[0024] System 10 performs three types of processing on signals from sensors 4. The in-cylinder pressure waveform analysis unit 20 analyzes the in-cylinder pressure waveform (pressure waveform inside the cylinder) for each cylinder of the engine 3. The in-cylinder pressure waveform differs depending on the engine load factor, and when the load factor is low and there is a lot of acceleration and deceleration, such as when entering or leaving port, the in-cylinder pressure waveform becomes unstable. Therefore, the memory unit 14 stores the normal in-cylinder pressure waveform (standard waveform) for each load factor of the engine 3 (for example, with a load factor range of ±0.5% in 1% increments) for each crank angle, for load factors in the range of, for example, 40% to 100%. The standard waveform is, for example, the waveform for each individual cylinder. When the load factor is, for example, less than 40%, the in-cylinder pressure waveform may be unstable, so the analysis of the in-cylinder pressure waveform is omitted. The load factor is the ratio of the output of engine 3 to the maximum continuous operating output, and can be estimated, for example, from the rotational speed of engine 3, or more precisely from the rotational speed of engine 3 and the rack value of the fuel injection pump.

[0025] The analysis unit 20 outputs the sum of the squares of the difference (error) between the measured in-cylinder pressure waveform (measured waveform) and the standard waveform, accumulated over one or more cycles of the engine 3's operation (a sum of values ​​representing the squared norm of the error). Instead of outputting the squared norm directly, the standard deviation of the normal squared norm may be calculated, and the measured squared norm divided by the standard deviation (error normalized by the standard deviation) may be output. Instead of simply outputting the squared norm of the error, errors for each crank angle may also be added.

[0026] The other sensor signal analysis unit 30 compares the signals from sensors other than the in-cylinder pressure sensor with a threshold value obtained by dividing the shift davg from the average value of the normal signal by the standard deviation σ, which is davg / σ. There are, for example, four types of thresholds: positive side warning, positive side predictive detection, negative side predictive detection, and negative side warning.

[0027] Even if a warning sign cannot be identified from the in-cylinder pressure waveform or from the signals of other individual sensors, it may be possible to identify it by observing the signals of multiple sensors as a sequenced set of data (vector data). For example, if the signals of 2 to 10 types of sensors fluctuate while maintaining their interrelationships, and the relationships between the sensor signals are different from normal, it may indicate some kind of warning sign regarding the engine 3 and its peripheral equipment. In other words, even if the signals of each individual sensor appear to fluctuate randomly within a narrow range, the combination of signals from multiple types of sensors may be fluctuating significantly. The relationship analysis unit 40 compares the relationships between the signals of multiple types of sensors (data from sets of sensor signals, i.e., vector data) with normal values ​​and detects warning signs of abnormality.

[0028] In relationship analysis, signals from sensors 4 are treated as vector data, and attention is paid to the fact that the vector data is shifted from the normal distribution. In this case, the vector data can be a large vector data (a single vector combining the signals from all sensors), or it can be a set of signals from multiple sensors that are presumed to be related based on empirical rules, extracted from the large vector data. Reducing the dimensionality of the vector makes processing easier. The memory unit 42 stores past data related to relationships, that is, data on which signals from which sensors are related, and within the range of relationships stored in the memory unit 42, it extracts the data within the range of relationships from the large vector data, creates small vector data, and stores its distribution.

[0029] The determination unit 50, based on the overall signals from the analysis units 20, 30, and 40, classifies the state of the engine 3 and its peripheral equipment into three stages, such as warning, precursor, and normal, and outputs this along with the sensor signal vector to the AI ​​60 for generating suggestions.

[0030] The AI ​​60 for generating suggestions refers to the distribution of past experiences (vector data of sensor signals and corresponding maintenance results) stored in the memory unit 62 and generates maintenance suggestions in response to the signal from the determination unit 50. The suggestions are estimated values ​​of the areas that need inspection and the condition of those areas. The output of the AI ​​60 is displayed on the monitor 70, edited by the operator, and transmitted via the memory unit 14 and communication unit 12 to the shipping company's land-based operation center 80, the ship 2, etc. Manual editing by the operator is not required.

[0031] On vessel 2, maintenance is performed on the engine 3 and other components based on instructions from the land-based operation center 80 or suggestions from system 10. At this time, the validity of the suggestions generated by system 10 is determined. The maintenance results from vessel 2 are then transmitted to system 10 either directly from the communication unit 7 or via the land-based operation center 80. The detection system 10 stores the maintenance suggestions actually transmitted by system 10, candidate suggestions generated by AI 60, the vector data of the sensors 4 at that time, and the actual maintenance results in the memory unit 14, memory unit 62, etc.

[0032] Figures 2 and 3 schematically show the types of sensors 4 and their signals. The main engine speed is data indicating the load of the engine 3. Sensors 4 include those that measure data for each cylinder, such as the in-cylinder pressure waveform sensor and the exhaust gas outlet temperature sensor in Figure 2, and those that measure data relating to the entire engine 3, such as the turbocharger exhaust gas outlet temperature sensor in Figure 2, which measures data without cylinder numbers. The types and number of sensors 4 are arbitrary, and in this embodiment there are approximately 300 types.

[0033] Figure 4 shows the in-cylinder pressure waveform for each cylinder over one cycle of engine 3. The horizontal axis represents the crank angle, and the vertical axis represents the in-cylinder pressure. In-cylinder pressure is basic data that indicates the state of each cylinder, and in addition to the fuel combustion state, it also indicates the state of the crank, fuel injection system, exhaust valve, intake valve, etc., and the sliding condition of the cylinder and piston.

[0034] In-cylinder pressure changes with engine load, and tends to be unstable at low loads and when load fluctuations are large. Therefore, the standard waveform and the measured waveform are compared for each load, and the in-cylinder pressure waveform is excluded from the diagnosis when the load is, for example, 40% or less. In addition, the diagnosis may be stopped when the engine load fluctuations are large. The standard waveform is the average value of past normal waveforms, and if data is insufficient, data from land tests is used as a substitute. Furthermore, if data for each load is insufficient, waveforms of surrounding loads are interpolated or extrapolated. The shift from the standard waveform is, in the embodiment, the cumulative value of the square of the shift for each crank angle, and when expressed as a vector, it is the square norm of the difference between the vector corresponding to the standard waveform and the measured vector.

[0035] The norm is not limited to the square norm; it can also be the product of the absolute values ​​of the differences (first-power norm), the maximum absolute value of the differences (infinity norm), etc. Furthermore, if it is necessary to increase the contribution of the high-pressure portion in the in-cylinder pressure waveform, the vector values ​​may be squared or cubed, the difference calculated, and then the product added. Additionally, if important parts of the in-cylinder pressure waveform are known, the weighting of the product in the product can be varied between the important parts and the other parts. In the example, the product of the differences in the in-cylinder pressure waveform is a single positive data point, but the in-cylinder pressure waveform may be divided for each crank angle range, resulting in what appears to be multiple data points.

[0036] Figure 5 shows the configuration of the abnormality prediction system 10. The main input signals are the cylinder pressure waveform and other sensor signals, with auxiliary inputs including temperature, the ship's current position, weather, wind speed and direction, waves, and smoke. Of these, weather, wind speed and direction, waves, and smoke are manually entered, for example, from the monitor 5 on board. The cylinder pressure waveform analysis unit 20 detects signs that could develop into abnormalities in the engine 3 and its related components through analysis of the cylinder pressure waveform. The standard waveform used for comparison is obtained by classifying and averaging the normal waveforms stored in the memory unit 14 according to the load factor, and if data is insufficient, it is supplemented by interpolation or extrapolation from the surrounding load factor.

[0037] The analysis unit 30 compares the individual sensor signals, other than the in-cylinder pressure waveform, with the average value and standard deviation of the signals under normal conditions. As shown in Figures 2 and 3, even the signals from the same sensor change depending on the load factor of the engine 3. Therefore, a correction is applied based on the load factor of the engine 3, and if there is insufficient data, the data at ambient load factors is interpolated using a third-order or sixth-order polynomial to supplement the data. Based on experience, interpolation using a third-order or sixth-order polynomial resulted in smooth interpolation of the sensor signals.

[0038] The analysis unit 30 classifies the sensor signals into, for example, five categories: a warning on the positive side (large sensor signal), a precursor on the positive side, normal, a precursor on the negative side, and a warning on the negative side (small sensor signal), based on the value obtained by dividing the shift from the normal value by the standard deviation. The analysis unit 30 converts the signal of each sensor into a data set consisting of the above five classifications and the value obtained by dividing the shift from the normal value by the standard deviation. Maintenance results from engine 3, etc., may reveal that even if the sensor signal has shifted only slightly from the normal value to the extent that it does not exceed a threshold, it may be a precursor to the need for maintenance. In such cases, it is preferable to reduce the threshold to detect even slight precursors.

[0039] The memory unit 42 attached to the relationship analysis unit 40 stores the distribution of past data related to relationships. This data includes which sets of components of the vector data (i.e., subvectors of the large vector) are related, and the relationship under normal conditions (distribution of the values ​​of the sensor signals with relationships = subvectors). In addition, for data that may be considered as a precursor to an anomaly during maintenance, etc., the memory unit 42 stores the type of sensor involved and the vector data (subvector data) corresponding to the precursor. Note that each component of the vector data is the signal of the individual sensor. The relationship analysis unit 40 calculates the difference between the data of the sets of sensor signals with relationships and the average value of the data under normal conditions, normalizes the calculated difference by dividing it by the standard deviation of the distribution under normal conditions, and outputs it to the determination unit 50.

[0040] The determination unit 50 makes a preliminary determination, based on the signals from the analysis units 20, 30, and 40, whether maintenance (inspection) is required for the engine 3 and its peripheral equipment. The signal from the analysis unit 30 has already been compared with a threshold, so it is output directly to the AI ​​60. The determination unit 50 evaluates the shift of the in-cylinder pressure waveform obtained by the analysis unit 20 and outputs the evaluation result to the AI ​​60. Similarly, the signal from the analysis unit 40 is compared with the acceptable range and output to the AI ​​60.

[0041] The reason for providing a determination unit 50 between AI60 and analysis units 20, 30, and 40 is to separate the processing of sensor signals in analysis units 20, 30, and 40 from the determination process, and to allow for flexible determination of whether processing by AI60 is necessary. Note that the determination unit 50 is not required. Furthermore, the threshold values ​​in the determination unit 50 are modified by AI60, and the determination unit 50 modifies the threshold values ​​for individual sensor signals in analysis unit 30.

[0042] AI60 generates maintenance suggestions for the shipping company's shore operations center or the ship's engine room personnel based on the data and judgment results from the judgment unit 50. The data from the judgment unit 50 consists of vector data from various sensors, or relevant parts of that data with supplementary data such as current position, ship speed, destination, weather, wind speed and direction, waves and smoke added.

[0043] The determination unit 50 considers any unusual data obtained from the sensor as a sign of an anomaly, so most of the detected signs are false alarms. The memory unit 62 stores vector data, determination results, and transmitted suggestions from past maintenance suggestions sent to the ship. In addition, it stores inspection results from when the engine or other parts of the ship were inspected in response to the suggestions. By comparing the determination result for an anomaly with past inspection results, more appropriate suggestions can be obtained. In this case, the degree of similarity of the vector data and past inspection results are taken into consideration.

[0044] The suggestions generated by AI60 are displayed on monitor 70 along with relevant vector data, etc. (Figure 1), and after modifications by the operator, they are transmitted to ship 2 via memory unit 14 and communication unit 12. Ship 2 uses the suggestions as a reference to perform inspections and repairs, and transmits the maintenance results to system 10 or land-based operation center 80, etc.

[0045] Figure 6 shows the processing of the in-cylinder pressure waveform. A standard in-cylinder pressure waveform corresponding to the engine load factor is prepared. If there are insufficient measured waveforms, waveforms at nearby load factors are interpolated or extrapolated. The square of the difference between the standard in-cylinder pressure waveform and the measured waveform is accumulated by the integration unit 22 for one or more cycles of engine operation to obtain the error in the in-cylinder pressure waveform (error expressed as a squared norm). The error may be weighted according to the crank angle, etc., and the norm may be a first-power norm, etc. In addition to the squared norm of the error, the error waveform may also be output.

[0046] Since the in-cylinder pressure waveform is a periodic waveform, it may be converted to a Fourier series using the in-cylinder pressure waveform analysis unit 25 in Figure 7. The Fourier converter 15 converts the standard in-cylinder pressure waveform for each load factor into a Fourier series. The measured in-cylinder pressure waveform is converted to a Fourier series by the Fourier converter 26, and the integration unit 27 integrates the squares of the errors for each Fourier component and outputs the square root. In addition, the Fourier transform of the errors may also be output. Furthermore, each component of the Fourier series may be multiplied by a weight for each component.

[0047] In analyzing the in-cylinder pressure waveform, instead of treating the entire waveform as a single data point, it may be divided into multiple data points based on factors such as the crank angle, and each data point may be analyzed separately. Alternatively, instead of using the error norm, the in-cylinder pressure waveform may be treated as vector data, and the angle between the standard waveform vector and the measured waveform vector may be calculated using the dot product. For example, if the standard waveform vector is a, the measured waveform vector is b, and their norms are |a| and |b|, then (a|b) / |a||b|≦1, and the degree to which it deviates from 1 represents the difference in the vectors.

[0048] Figure 8 schematically shows vector data 100 in an embodiment. Data 100 includes in-cylinder pressure waveform and other sensor data, date and time, destination, ship speed, weather, temperature, etc. The required parts are extracted from this and used in analysis units 20, 30, and 40. Alternatively, a time series of vector data 100 may be generated and the time series data may be used as the target of analysis. AI 60 then generates suggestions based on the similarity between standard vector data and measured vector data, and past inspection results. This eliminates false alarms. Furthermore, based on the above similarity and past inspection results, the thresholds in the judgment unit 50 and analysis unit 30 are adjusted to better avoid missing warning signs. This makes it possible to detect warning signs that are very close to the normal state.

[0049] Figure 9 shows the processing of sensor signals other than the in-cylinder pressure waveform. The analysis unit 30 reads standard waveforms of past sensor signals, organized by engine load factor, from the storage unit 14. If there are insufficient standard waveforms at the same load factor as the current engine load factor, data at surrounding load factors is interpolated or extrapolated. A cubic or sixth-degree polynomial of the load factor is suitable for interpolation or extrapolation, and it has been empirically found that a cubic polynomial is particularly suitable.

[0050] The error extraction unit 31 calculates the absolute value of the error between the standard waveform and the measured waveform, and the comparison unit 32 compares it with thresholds stored in the storage unit 33. There are four types of thresholds, for example, alarm thresholds for both positive and negative sides, and preventive maintenance thresholds for both positive and negative sides. The comparison results with the thresholds (four types in addition to normal) are output to the determination unit 50. The AI ​​60 also changes the thresholds.

[0051] Figure 10 shows the relationship analysis algorithm. The relationship analysis unit 40 analyzes the behavior of signals between multiple sensors and extracts those that deviate from past normal behavior. When focusing on individual sensor signals, they may fall within the normal range, but when focusing on combinations of signals from multiple sensors, movements that do not occur under normal conditions may be detected. In this way, warning signs that cannot be detected from individual sensor signals are detected. If data from when a warning sign exists is available, the degree to which it is close to the data indicating the warning sign can also be used. The analysis unit 40 then generates thresholds for warnings and warning sign detection in the relationship analysis from the distribution of past sensor signal combinations. The output unit 44 outputs the signals of the relevant sensors and the degree to which they exceed the threshold, and the determination unit 50 makes a determination.

[0052] Although a diesel engine was used as an example in the embodiment, the fact that the in-cylinder pressure waveform contains a lot of data about combustion in the cylinder is also common to Otto engines. Furthermore, the ability to detect signs of abnormality in marine engines from sensor signals other than the in-cylinder pressure waveform is also the same for Otto engines. [Explanation of Symbols]

[0053] 2 ships 3. Diesel engine (main engine) 4. Sensors 5 Computers 6 monitors 7 Communications Department 10. Anomaly detection system 12 Communications Department 14 Storage section 15,26 Fourier Transformer 20,25 In-cylinder pressure waveform analysis section 22,27 Estimation Department 30 Other Sensor Signal Analysis Unit 31 Error extraction part 32 Comparison Section 33. Memory Unit (Alarm threshold and Anomaly Prediction Detection Threshold) 40 Relationship Analysis Department 42. Memory section (past data related to relationships) 44 Output section 50 Judgment section 60 AI for generating suggestions 62 Memory section (past knowledge, rules and materials to provide insights, etc.) 70 monitors 80 Land Operations Center 100 vector data

Claims

1. Equipped with a computer for detecting signs of abnormality in marine engines, We receive data related to marine engines. The aforementioned computer, A storage unit that stores standard waveforms showing typical in-cylinder pressure waveforms for marine engines for each load factor of the marine engine, An in-cylinder pressure waveform analysis unit uses a measured waveform showing the in-cylinder pressure waveform of a marine engine and a standard waveform read from the storage unit according to the load factor of the marine engine, and outputs data showing the integrated absolute value of the difference between the measured waveform and the standard waveform for at least one cycle of the operation of the marine engine. The in-cylinder pressure waveform analysis unit compares the integrated data output by the unit with a threshold value and functions as a determination unit to detect signs of abnormality in the marine engine. The aforementioned storage unit also stores the distribution of past data for multiple types of data other than the in-cylinder pressure waveform related to marine engines. The aforementioned computer generates thresholds for each of the multiple types of data from the distribution of past data, compares the thresholds with the current data, and detects signs of abnormality in the marine engine or its peripheral equipment according to the type of data that exceeds the threshold, thereby providing a marine engine abnormality prediction device.

2. The in-cylinder pressure waveform analysis unit is characterized in that it integrates the absolute values ​​of the differences in the in-cylinder pressure waveforms for each crank angle and uses the integrated value as data, thereby providing a device for detecting abnormal signs in a marine engine according to claim 1.

3. The computer also stores the distribution of past data regarding a predetermined combination of data among the multiple types of data. A device for detecting signs of abnormality in a marine engine or its peripheral equipment, characterized in that it detects signs of abnormality in a marine engine or its peripheral equipment by performing relationship analysis on the current data with respect to a predetermined combination of data, based on the distribution of stored past data.

4. The aforementioned memory unit stores data from when abnormal signs were detected in the past, and the inspection results of the marine engine at that time. The marine engine abnormality detection device according to claim 3, characterized in that the computer refers to data from past abnormality detections and inspection results with respect to similar abnormality signs, and generates suggestions regarding inspections of the marine engine or its peripheral equipment.

5. The device for detecting abnormal signs in a marine engine according to claim 4, characterized in that the computer refers to past inspection results regarding similar abnormal signs, including results indicating that maintenance is necessary and results indicating that maintenance is not necessary, and reduces the need for maintenance in the indication when the results indicating that maintenance is not necessary are predominant.

6. The computer stores a threshold for detecting abnormal signs based on data showing the in-cylinder pressure waveform, the multiple types of data, and the predetermined data combination. The marine engine abnormality detection device according to claim 5, characterized in that, based on the result obtained during an inspection when an abnormality is detected that maintenance is necessary, the threshold is lowered.

Images