A maintenance demand prediction method based on ship equipment maintenance history records

Abstract

The present invention discloses a maintenance demand prediction method based on the maintenance history record of ship equipment. First, a statistical time unit is set, and the ship maintenance history data is statistically collected with the statistical time unit as the minimum unit; then data cleaning is performed to clean the ship maintenance history data to obtain standardized data; then a prediction model is established, including a non-homogeneous Poisson process model and a geometric process model; the standardized data of the previous statistical time unit is substituted into the prediction model for training to obtain the prediction model parameters, and the maintenance demand of several subsequent statistical time units is predicted. The present invention proposes a new ship maintenance demand prediction method, which can achieve accurate prediction in the face of the complexity of ship system equipment, the diversity of use environment and mission profiles, and the inaccuracy of data collection, thereby improving the reliability of ship maintenance guarantee during navigation, reducing maintenance costs, and preventing ocean-going ships from carrying too many useless spare parts.

Description

Technical Field

[0001] The invention belongs to the field of ship maintenance and support, relates to the historical maintenance workload of ship equipment, and specifically is a maintenance demand prediction method based on the historical maintenance records of ship equipment. Background Art

[0002] The existing methods for predicting the maintenance demand of ship equipment mostly use traditional reliability data analysis methods mainly for equipment failure interval time data (Time Between Failures Data, TBF). This type of data type is more suitable for data analysis of component-level equipment in the development and testing stage. For the maintenance workload analysis and prediction work in the use stage of ship equipment, due to the complexity of ship system equipment, the diversity of use environment and mission profiles, and the inaccuracy of data collection, the use of failure count data has better fault tolerance and robustness. Therefore, a method for predicting maintenance demand based on ship failure count data type is proposed. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the existing technology, to analyze and predict the maintenance workload during the use stage of ship equipment, and to propose a maintenance demand prediction method based on the historical records of ship equipment maintenance, so as to use the maintenance data of the historical records to predict future maintenance needs and improve the reliability and economy of ship maintenance guarantee.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0005] In one aspect, the present invention provides a maintenance demand prediction method based on ship equipment maintenance history records, comprising the following steps:

[0006] Setting a statistical time unit, using the statistical time unit as the minimum unit to count the ship maintenance history data, the ship maintenance history data including the maintenance process record data of all equipment failures on the ship;

[0007] Data cleaning: clean the historical data of ship maintenance, correct irregular records, and obtain standardized data;

[0008] Establishing a prediction model, the prediction model includes a non-homogeneous Poisson process model or a geometric process model;

[0009] The standardized data of historical continuous statistical time units are substituted into the prediction model for training, the model parameters of the corresponding prediction model are obtained, and the maintenance requirements of subsequent statistical time units are predicted.

[0010] Preferably, the statistical time unit includes day, month, quarter and year. Considering the periodicity and uncertainty of ship missions, if the counting time unit is too short, the prediction result will not be too accurate, and if the time is too long, it will not conform to the actual situation. Therefore, for offshore ships, months are generally used as the unit, and for ships mainly engaged in ocean voyages, years are selected as the counting time period.

[0011] Preferably, in the process of counting historical ship maintenance data, the accumulated working time of each faulty equipment is counted, and when the statistical time unit is set, the time unit is greater than the single longest running time of all equipment.

[0012] Preferably, the data cleaning includes deleting duplicate information, correcting erroneous information (such as illegal dates), merging similar items, and filling in incomplete data entries. The original historical maintenance data records are cleaned to make the data as regular as possible to facilitate the next step of data statistics and analysis.

[0013] Preferably, the prediction model is a non-homogeneous Poisson process model (NHPP), and the prediction model is defined as follows:

[0014] Assume that the maintenance workload X(t) occurring before time t is an independent incremental process, and define the maintenance workload X(t) as Non-homogeneous Poisson process, λ(t) is the intensity function of maintenance workload;

[0015] According to the expectation of the nonhomogeneous Poisson distribution is the expected maintenance workload corresponding to time t;

[0016] Since λ(t) is the interval between maintenance work and (i=1.2.3….n, n is the total number of time periods) The same distribution obeys the exponential distribution, so the classic exponential model is used, and its intensity function is defined as:

[0017] Formula (1)

[0018] The parameter λ represents the initial strength, that is, the initial statistical maintenance workload, and β is a parameter representing the strength growth rate. For the exponential model, the strength change trend can be known by judging the parameter β. Obviously, when β is greater than 1, the system has a tendency to increase failures, and when β is less than 1, the surface failures tend to decrease. For β=1, the exponential model degenerates into a homogeneous Poisson process model;

[0019] By counting the number of maintenance workloads in different time periods with a uniform fixed duration, the number of maintenance workloads in the following time periods, i.e. the expected maintenance workload, is predicted. For the convenience of mathematical processing, the first i Statistics of maintenance times in a certain interval X i As The estimated value of is:

[0020] Formula (2)

[0021] Indicates i The initial intensity of maintenance workload in a period of time, represents the strength growth rate coefficient in the i-th time period, express i The length of the time period;

[0022] Using the least squares method, the estimated values ​​of the parameters λ and β for the n+1th time period are calculated using the standardized data of the previous n time periods. and :

[0023] Formula (3)

[0024] Formula (4)

[0025] Maintenance workload forecast for the next time period: Substitute the estimated values ​​of parameters λ and β and Substitute into formula (2) to calculate the maintenance workload in time period n+1.

[0026] Preferably, the prediction model may also be a geometric process model (GP), and the prediction model is defined as follows:

[0027] Definition 1 (Geometric process): If a sequence of non-negative random variables { Y , n = 1, 2, 3…} constitutes an update process, then for a given real number α, the random variable sequence { Xn = α n-1 Y , n = 1, 2, 3…} is called a geometric process.

[0028] Obviously, the geometric process can be intuitively understood as a "geometric random sequence". Xn Indicates nThe maintenance workload of each interval, then the parameter α is the proportional coefficient of the increase in the mean workload. If α = 1, the geometric process is a classic update process, reflecting that the system is in a stable period and the workload intensity has not increased over time; if α>1, the mean value of Xn increases geometrically, indicating that the workload intensity increases; and α<1 indicates that the workload intensity tends to decrease. Since the mathematical description of the geometric process is concise and the physical meaning of the parameters is clear, it is conducive to the quantitative analysis of interval counting type data.

[0029] use X i Indicates i The maintenance workload in a time interval, then the parameter α is the proportional coefficient of the mean increase in workload; for the parameter estimation problem of geometric process, non-parametric method is used for estimation, let

[0030] Formula (5)

[0031] n is the total number of time periods, For the i A non-negative random variable for a time period;

[0032] Taking the logarithm of both sides of formula (5) we have

[0033] Formula (6)

[0034] because Y i It constitutes an update process, so let , , , then formula (6) can be written as

[0035] Formula (7)

[0036] E represents expectation. In formula (7), is a set with mean 0 and variance A sequence of random variables;

[0037] The prediction model is trained as follows:

[0038] The linear regression method is used to calculate the estimated values ​​of α and λ for formula (7), as shown in formulas (8) and (9):

[0039] Formula (8)

[0040] Formula (9)

[0041] Using the historical maintenance records of the previous n time periods, substitute them into formula (8) and formula (9) to obtain the estimated values ​​of α and λ for the n+1 time period: and ;

[0042] Prediction of maintenance workload in the next time period: Substitute the estimated values ​​of α and λ in time period n+1 into formula (7) to predict the maintenance workload in time period n+1.

[0043] Preferably, at least n+1 time periods of historical ship maintenance data are counted, and the maintenance workload in n+1 time periods is predicted using a prediction model using the standardized data of the previous n time periods. The predicted maintenance workload is compared with the statistical maintenance workload. If the error meets the requirements, it means that the prediction model is reasonable.

[0044] Preferably, multiple models can be used to predict the maintenance workload in n+1 time periods, and the model with high accuracy is selected after testing to predict the maintenance workload in n+1 time periods.

[0045] On the other hand, the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned maintenance demand prediction method when executing the program.

[0046] On the other hand, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the above-mentioned maintenance demand prediction method when executed by a processor.

[0047] On the other hand, the present invention provides a computer program product, including a computer program, which implements the above-mentioned maintenance demand prediction method when executed by a processor.

[0048] On the other hand, the present invention also provides a maintenance demand prediction system based on the maintenance history records of ship equipment, comprising:

[0049] Data statistics module, used to count historical data of ship maintenance;

[0050] Data cleaning module, used to clean and standardize ship maintenance history data;

[0051] The prediction calculation module has a built-in prediction model, which is used to predict the maintenance needs of the n+1 time period based on the statistical standardized data of the previous n time periods.

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

[0053] The present invention proposes a new method for predicting ship maintenance needs, which can achieve accurate prediction in the face of the complexity of ship system equipment, the diversity of usage environment and mission profiles and the inaccuracy of data collection, thereby improving the reliability of maintenance support during the ship's navigation process, reducing maintenance costs and preventing ocean-going ships from carrying too many useless spare parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The figure is a flow chart of a maintenance demand prediction method based on ship equipment maintenance history records of the present invention.

[0055] Figure 2 It is a stacked bar chart of the maintenance workload of a certain type of ship equipment.

[0056] Figure 3 It is the fitting result of the non-homogeneous Poisson process (total workload and host group).

[0057] Figure 4 is the geometric process fitting result (boiler group and refrigeration group). DETAILED DESCRIPTION

[0058] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0059] like Figure 1 As shown, the present invention provides a maintenance demand prediction method based on ship equipment maintenance history records, comprising the following steps:

[0060] S100, setting a statistical time unit, using the statistical time unit as the minimum unit to collect historical ship maintenance data, wherein the historical ship maintenance data includes the maintenance process record data of all equipment failures on the ship;

[0061] The statistical time units include day, month, quarter and year. Considering the periodicity and uncertainty of ship missions, if the counting time unit is too short, the prediction result will not be too accurate, and if the time is too long, it will not conform to the actual situation. Therefore, for offshore ships, months are generally used as the unit, and for ocean-going ships, years are selected as the counting time period.

[0062] It should be noted that, in the process of statistically analyzing the historical data of ship maintenance, the cumulative working time of each faulty equipment is counted, and when the statistical time unit is set, the time unit is greater than the single longest operating time of all equipment.

[0063] All equipment includes control systems, supply devices, boilers, turbine circulating water pumps, power stations, turbine generators, pipelines, valves and other ship operating equipment and ancillary equipment.

[0064] S200, data cleaning, cleaning the historical data of ship maintenance, correcting irregular records, and obtaining standardized data;

[0065] The data cleaning includes deleting duplicate information, correcting erroneous information (such as illegal dates), merging similar items, and filling in incomplete data entries. The original historical maintenance data records are cleaned to make the data as regular as possible to facilitate the next step of data statistics and analysis.

[0066] S300, establishing a prediction model, where the prediction model includes a non-homogeneous Poisson process model or a geometric process model;

[0067] The prediction models include two types, namely, Non Homogeneous Poisson Process Model (NHPP) and Geometric Process Model (GP).

[0068] S310. For the Non Homogeneous Poisson Process (NHPP) model, the model definition is as follows: Assume that the maintenance workload X(t) occurring before time t is an independent incremental process, and the maintenance workload X(t) is determined to obey Non-homogeneous Poisson process, λ(t) is the intensity function of maintenance workload;

[0069] According to the expectation of the nonhomogeneous Poisson distribution is the expected maintenance workload corresponding to time t;

[0070] Since λ(t) is the interval between maintenance work and (i=1.2.3…n, n is the total number of time periods) The same distribution obeys the exponential distribution, so the classic exponential model is used, and its intensity function is defined as:

[0071] Formula (1)

[0072] The parameter λ represents the initial strength, that is, the initial statistical maintenance workload, and β is a parameter representing the strength growth rate. For the exponential model, the strength change trend can be known by judging the parameter β. Obviously, when β is greater than 1, the system has a tendency to increase failures, and when β is less than 1, the surface failures tend to decrease. For β=1, the exponential model degenerates into a homogeneous Poisson process model;

[0073] By counting the number of maintenance workloads in different time periods with a uniform fixed duration, the number of maintenance workloads in the following time periods, i.e. the expected maintenance workload, is predicted. For the convenience of mathematical processing, the first i Statistics of maintenance times in a certain intervalX i As The estimated value of is:

[0074] Formula (2)

[0075] Indicates i The initial intensity of maintenance workload in a period of time, represents the strength growth rate coefficient in the i-th time period, express i The length of the time period;

[0076] The prediction model is trained as follows:

[0077] Parameter estimation for the next time period: Using the least squares method, the estimated values ​​of the parameters λ and β for the n+1th time period are calculated using the standardized data of the previous n time periods. and :

[0078] Formula (3)

[0079] Formula (4)

[0080] S320, for the Geometric Process (GP) model, the model definition is as follows:

[0081] Definition 1 (Geometric process): If a sequence of non-negative random variables { Y , n = 1, 2, 3…} constitutes an update process, then for a given real number α, the random variable sequence { Xn = α n-1 Y , n = 1, 2, 3…} is called a geometric process.

[0082] Obviously, the geometric process can be intuitively understood as a "geometric random sequence". Xn Indicates n The maintenance workload of each interval, then the parameter α is the proportional coefficient of the increase in the mean workload. If α = 1, the geometric process is a classic update process, reflecting that the system is in a stable period and the workload intensity has not increased over time; if α>1, the mean value of Xn increases geometrically, indicating that the workload intensity increases; and α<1 indicates that the workload intensity tends to decrease. Since the mathematical description of the geometric process is concise and the physical meaning of the parameters is clear, it is conducive to the quantitative analysis of interval counting type data.

[0083] use X i Indicates i The maintenance workload in a time interval, then the parameter α is the proportional coefficient of the mean increase in workload; for the parameter estimation problem of geometric process, non-parametric method is used for estimation, let

[0084] Formula (5)

[0085] n is the total number of time periods, For the i A non-negative random variable for a time period;

[0086] Taking the logarithm of both sides of formula (5) we have

[0087] Formula (6)

[0088] because Y i It constitutes an update process, so let , , , then formula (6) can be written as

[0089] Formula (7)

[0090] E represents expectation. In formula (7), is a set with mean 0 and variance A sequence of random variables;

[0091] The prediction model is trained as follows:

[0092] Parameter estimation for the next time period: The linear regression method is used to calculate the estimated values ​​of α and λ using formula (7), as shown in formulas (8) and (9):

[0093] Formula (8)

[0094] Formula (9)

[0095] Using the historical maintenance records, we can substitute them into formula (8) and formula (9) to obtain the estimated values ​​of α and λ for the n+1 period: and ;

[0096] S400: Substitute the standardized data based on the statistical time unit into the prediction model for training, obtain the prediction model parameters, and predict the maintenance needs of several subsequent statistical time units.

[0097] S410. For the non-homogeneous Poisson process model, the estimated values ​​of the parameters λ and β in the n+1 time period are used. and Substitute into formula (2) to calculate the maintenance workload in time period n+1.

[0098] S420. For the geometric process model, the estimated values ​​of α and λ in the n+1 time period are substituted into formula (7) to predict the maintenance workload of n+1.

[0099] At least n+1 time periods of historical ship maintenance data can be counted, and the standardized data of the previous n time periods can be used to adopt a prediction model to predict the maintenance workload of n+1 time periods. The predicted maintenance workload is compared with the statistical maintenance workload. If the error meets the requirements, it means that the prediction model is reasonable.

[0100] Verification case: A case analysis was conducted on the maintenance record information collected from a steam-powered ship over a continuous period of 5 years to verify the feasibility of the demand prediction technology based on the maintenance record information.

[0101] Step 1: Prepare data. The maintenance data collection results of a certain ship are provided in the form of Excel spreadsheets, some data samples of which are shown in Table 1.

[0102] Table 1 shows some maintenance data of a ship

[0103] Serial number major Failure time Repair Nature Equipment Name Fault phenomenon Treatment measures Spare part name Consumption quantity Spare parts source Repair unit (person) Exclude time Remark 1 Electromechanical January 4 Self-study Supply device Internal seal ring broken Replace the seal ring 1 Oil Group 2 Electromechanical January 8 Factory Repair Main boiler No. 3, main boiler No. 4 The speed of the THA fan of the No. 3 main boiler is higher than that of No. 4 at high load On February 23, it was found that the Oma stroke of the No. 3 THA fan was 20 mm, and the Oma stroke of the No. 4 YHA fan was 25 mm. Boiler Group 3 Electromechanical January 8 Self-study 2# Steam turbine circulating water pump Abnormal noise occurs in the pump body Shut down the machine, close the seabed door, open the main condenser manhole, crawl in and open the anti-corrosion zinc block of the pump body for inspection Host Group 2 January 8 4 Electromechanical January 8 Factory Repair Former power station Display screen is distorted, no display If the monitor is disassembled and used on other computers and the screen is still distorted or black, it is considered to be a monitor failure. Check for third-level spare parts. Weak current group 5 Electromechanical January 8 Factory Repair 27# channel air conditioning is difficult to get sea water pipe 27# channel air conditioning is difficult to get sea water pipe Damaged and leaking Disassemble and inspect, there are sand holes in the flange, return to the factory for repair welding Water Group January 9 6 Electromechanical January 8 Factory Repair 1# Steam Turbine Generator High pressure drain leak Dismantle and return to the factory for grinding and repair Auxiliary unit January 10 7 Electromechanical January 8 Factory Repair 1# Turbine Generator Auxiliary Boiler Superheated Steam Valve External leakage, internal leakage Dismantle, inspect, replace gasket, and grind valve surface Auxiliary unit January 10 8 Electromechanical January 8 Factory repair, assisted repair 1# Turbogenerator Seal Saturated Steam Valve Internal leakage Disassembly and Grinding Auxiliary unit January 10 9 Electromechanical January 8 Factory Repair 1# Steam Turbine Generator Pressure Reducing Valve External leakage, internal leakage Dismantle, inspect, replace gasket, and grind valve surface Auxiliary unit January 10 10 Electromechanical January 1 Factory Repair 1# auxiliary condenser auxiliary injection new steam threshold Internal leakage Disassembly and Grinding Auxiliary unit January 10 11 Electromechanical January 11 Self-study Pumping system Valve internal leakage Replacement valve GB596-83 External thread stop valve 1 base Auxiliary boiler group January 10 12 Electromechanical January 12 Self-study 5# feces cabinet ejector Damaged water Replacement Water Group January 12 13 Electromechanical January 16 Factory Repair No. 3 diesel generator Diesel filter venting air plugs oil leakage There is one more marble inside, the plug ejector is deformed, and the ejector needs to be re-processed Diesel Generator January 17 14 Electromechanical January 16 Factory Repair No. 2 water supply unit The cranking did not work at first, but later there was a clicking sound when the water pump was cranked. Dismantling and inspection on the 7th (disconnecting the reducer coupling and taking the entire feed pump out of the cabin for repair), it was taken out of the cabin on the 18th, and the workshop repaired it successfully in the evening of the 18th Sealing ring 1 Factory processing Host Group January 21 15 Electromechanical January 16 Factory Repair Starboard Captain's Boat Diesel engine heating plate does not work The first cylinder heater is damaged, the rest are normal Diesel Generator 16 Electromechanical January 17 Factory Repair No. 4 diesel generator When paralleling, load grabbing occurs and reactive current is large Replacement of voltage regulator parts BKH-6B voltage corrector parts 1 Diesel Generator

[0104] Step 2: Build a data cleaning tool to review and verify the original data, delete duplicate information, correct erroneous information (such as illegal dates), merge similar items, and fill in incomplete data items as appropriate. A sample table of some data after cleaning is shown in Table 2. Based on the cleaned data, the workload of each repair unit for the ship for five consecutive years was counted, as shown in Table 3.

[0105] Table 2 shows the maintenance data of a certain ship after cleaning

[0106] Serial number System Name Fault phenomenon Treatment measures Spare part name Failure time Recovery time Maintenance time (hours) Repair unit Repair Nature 1 Automatic control system Uncategorized other Coil one 7.26 7.26 0 Self-control group Self-study 2 Automatic control system Uncategorized other repair 8.23 8.23 0 Self-control group Self-study 3 Host leakage Grinding 9.29 9.29 0 Host group Self-study 4 Host Uncategorized replace 9.24 9.24 0 Host group Self-study 5 Host damage replace Steam arrester 8.13 8.13 0 Host group Self-study 6 Host leakage replace O-Ring 9.1 9.1 0 Host group Self-study 7 Host leakage Grinding 9.26 9.29 72 Host group Factory repair (assisted repair) 8 Host leakage replace 8.2 8.2 0 Host Group 2 Self-study 9 Host leakage replace 8.21 8.21 0 Host Group 2 Self-study 10 Host Uncategorized replace O-Ring 9.21 9.21 0 Host Group 2 Self-study 11 Host Uncategorized replace O-Ring 7.3 7.3 0 Host Group 2 Self-study 12 Host leakage Welding repair 9.24 9.29 120 Host Group 2 Factory repair (assisted repair) 13 Host damage Welding repair 8.11 9.14 816 Host Group 2 Factory repair (assisted repair) 14 Host leakage Grinding 8.11 9.14 816 Host Group 2 Factory repair (assisted repair) 15 other Uncategorized Demolition and repair 9.26 9.29 72 Refrigeration Group Factory repair (assisted repair) 16 other Uncategorized Adjustment 7.15 7.18 72 Refrigeration Group Self-study 17 other Uncategorized Adjustment 7.15 7.18 72 Refrigeration Group Self-study 18 other Switch failure Grinding 8.5 8.6 24 Refrigeration Group Self-study 19 Oil system leakage Grinding 7.27 7.29 48 Oil Group Self-study 20 Oil system leakage replace 8.17 8.18 24 Oil Group Self-study 21 Oil system Uncategorized Grinding 8.26 8.27 24 Oil Group Self-study 22 Waterway Tie leakage replace 7.17 7.19 48 Water Group Self-study 23 Water system Cracking Welding repair 7.28 7.3 48 Water Group Self-study 24 Water system Uncategorized clean 8.15 8.18 72 Water Group Self-study 25 Water system Uncategorized replace Disc and check valve 8.11 8.12 24 Water Group Factory repair (assisted repair) 26 Water system Uncategorized clean 9.26 9.3 96 Water Group Factory repair (Assisted repair 27 Water system leakage replace 8.9 8.9 0 Water Group Self-study

[0107] Note: In Table 2, maintenance time 0 means that the repair is completed within one day.

[0108] Table 3 Workload statistics of each repair unit for a certain ship for 5 consecutive years

[0109]

[0110] Step 3: Data visualization analysis. Through data visualization, the changing trend of maintenance workload over time can be intuitively displayed. First, the total workload and the main engine class, boiler class and refrigeration class with heavier maintenance workload are selected. It can be seen from Table 3 that both the total workload and the workload of each major maintenance unit have a more obvious increasing trend. If a change curve chart is used, there is a disadvantage that it is not convenient to show the workload changes of all maintenance units. If displayed at the same time, it will lead to confusing display, which is not conducive to observation and loses intuitiveness. For the visualization of maintenance workload, a stacked bar chart is a better visualization method, such as Figure 1 The stacked bar chart clearly shows the changing trend of the overall equipment maintenance workload, and also intuitively shows the workload proportion of each maintenance unit. For example, from the figure, it can be clearly seen that the boiler team has the largest maintenance workload, while the workload of the refrigeration team, main engine team, automatic control team, water team, and diesel engine team also accounts for a large proportion.

[0111] Step 4: Two statistical models are used to analyze and process historical maintenance data, namely the NonHomogeneous Poisson Process (NHPP) model and the Geometric Process (GP) model.

[0112] Therefore, in this case, by substituting the data of the first four years in Table 3, the total workload data can be calculated. Using the non-homogeneous Poisson process model algorithm, the parameters are estimated. =45.201, parameters =1.409, substitute it into formula (2) to calculate, , get the number of maintenance times X i =251; using geometric process model algorithm, parameters = 64.798, parameters =1.404, substituted into formula (7), we get. Therefore, the fitting curve obtained by the geometric process is almost consistent with the result of the non-homogeneous Poisson process. The result of the prediction for the fifth year is 252, which is slightly better than the prediction result of the non-homogeneous Poisson process, 251.

[0113] Figure 3 The fitting results of the non-homogeneous Poisson process of the total maintenance workload and the maintenance workload of the main engine shift are given. The dotted line in the figure is the fitting result based on the data of the first four years using the above formula. Figure 3 It can be seen that the prediction results of the fitting formula for the total workload and the main engine class workload are relatively accurate. For the total workload, the predicted result is 251, while the actual result is 262; for the main engine class, the predicted result is 27, while the actual result is 29. It can be seen that the predicted results are in good agreement with the actual results.

[0114] Figure 4 The fitting results of the geometric processes of boiler class and refrigeration class are given. Figure 4 It can be seen that the prediction for the refrigeration class based on the fitting formula is relatively accurate, but there is a large error for the boiler class, with the predicted result being 48 and the actual result being 32. This error is caused by a variety of randomness, such as the randomness of fault occurrence and the randomness of task working time.

[0115] In summary, the statistical theory maintenance workload prediction method based on maintenance records has good accuracy overall, especially in predicting the maintenance workload change trend. The maintenance workload change trend curves of each maintenance unit remain consistent.

[0116] Due to the inherent random nature of maintenance demand, predictions for a shorter period of time (such as a month) may be very inaccurate, but predictions for a longer period of time (year) can reduce randomness and achieve better accuracy. In addition, the accuracy of predictions for different maintenance units is different, and some units' prediction results have certain errors (such as boiler shifts), which may be related to the working hours of each device. In order to further improve the accuracy of the analysis, it is recommended that the cumulative working time of the equipment at the time of the failure be collected synchronously in the future maintenance record collection system, so as to reduce the impact of prediction errors caused by different working hours in each year.

[0117] The above embodiments are only used to illustrate the present invention, but not to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that various combinations, modifications or equivalent substitutions of the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and should be included in the scope of the claims of the present invention.

Claims

1. A maintenance demand prediction method based on ship equipment maintenance history records, characterized in that: The following steps are involved: Setting a statistical time unit, using the statistical time unit as the minimum unit to count the ship maintenance history data, the ship maintenance history data including the maintenance process record data of all equipment failures on the ship; Data cleaning: clean the historical data of ship maintenance, correct irregular records, and obtain standardized data; A prediction model is established, which is a non-homogeneous Poisson process model. The prediction model is defined as follows: Assume that at time t Maintenance work done before X ( t ) is an independent incremental process that defines the maintenance workload X ( t ) to obey Non-homogeneous Poisson process, λ ( t ) is the intensity function of maintenance workload; According to the expectation of the nonhomogeneous Poisson distribution For the corresponding t The amount of maintenance work expected at any given time; The intensity function is defined using the exponential model as follows: Formula (1) The parameter λ represents the initial strength, i.e. the initial statistical maintenance workload, and β is a parameter representing the strength growth rate; For the exponential model, when β When it is greater than 1, it indicates that there is a trend of increasing faults. β When it is less than 1, it indicates that the faults have a decreasing trend. β =1, the exponential model degenerates into a homogeneous Poisson process model; By counting the number of maintenance workloads in different time periods with a uniform fixed duration, the number of maintenance workloads in the subsequent time periods, i.e. the expected maintenance workload, is predicted. i Statistics of maintenance times in a certain interval X i As The estimated value of is: Formula (2) Indicates i The initial intensity of maintenance workload in a period of time, represents the strength growth rate coefficient in the i-th time period, t i express i The length of the time period; The prediction model is trained as follows: Using the least squares method, n The standardized data of the time period is calculated n The estimated values ​​of the parameters λ and β for the +1 period: Formula (3) Formula (4) Maintenance workload forecast for the next time period: Substitute the estimated values ​​of parameters λ and β , Substituting into formula (2) we can get n +1 time period of maintenance work; The standardized data of historical continuous statistical time units are substituted into the prediction model for training, the model parameters of the corresponding prediction model are obtained, and the maintenance requirements of subsequent statistical time units are predicted.

2. The maintenance demand prediction method based on the ship equipment maintenance history record according to claim 1 is characterized by: The statistical time units include day, month, quarter and year.

3. The maintenance demand prediction method based on the ship equipment maintenance history record according to claim 1 is characterized by: The data cleaning includes deleting duplicate information, correcting erroneous information, merging similar items, and filling in incomplete data entries.

4. The maintenance demand prediction method based on the ship equipment maintenance history record according to claim 1 is characterized by: At least n+1 time periods of historical ship maintenance data are collected, and the maintenance workload of n+1 time periods is predicted using a prediction model using the standardized data of the previous n time periods. The predicted maintenance workload is compared with the statistical maintenance workload. If the error meets the requirements, it means that the prediction model is reasonable.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the maintenance demand prediction method according to any one of claims 1 to 4 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the maintenance demand prediction method according to any one of claims 1 to 4 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the maintenance demand prediction method according to any one of claims 1 to 4 is implemented.

Images