A mud pump life modeling method based on multi-source residual collaborative learning

Abstract

The present application relates to the technical field of dredging equipment maintenance, and particularly relates to a mud pump life modeling method based on multi-source residual collaborative learning, which comprises the following steps: firstly, acquiring simulation model output results simulating the work of a mud pump; then, collecting multi-source data of a mud pump work site; then, performing residual calculation according to the multi-source data and the simulation model output results to obtain multi-source residuals; and finally, correcting a preset mud pump life model based on the multi-source residuals to obtain an actual mud pump life model. The present application comprehensively covers the physical nature of the multi-factor coupling damage of the mud pump through multi-source data fusion, accurately identifies the deviation between the model assumption and the actual environment by using the residual-driven model correction mechanism, continuously corrects the mud pump life model based on real-time residuals, has dynamic iterative evolution capability, provides core support with mechanism depth and engineering practicability for the intelligent operation and maintenance of the dredging mud pump, and solves the problem of inaccurate mud pump life modeling in the prior art.

Description

Technical Field

[0001] This invention relates to the field of dredging equipment maintenance technology, and in particular to a mud pump life modeling method based on multi-source residual collaborative learning. Background Technology

[0002] As a core component of the high-concentration sediment transport system in dredging and propagation, mud pumps are prone to problems such as wear of key friction pairs, shaft imbalance, and seal fatigue during long-term operation. If the performance degradation trend cannot be predicted in time, it may lead to serious transport interruption or structural failure accidents.

[0003] In practice, unlike mud pumps in other fields, the lifespan degradation of dredging vessel mud pumps is the result of multiple coupled factors. For example, seawater corrosion reduces material hardness, ship swaying causes uneven clearance between the impeller and pump casing, leading to backflow wear, and frequent start-stop cycles cause the shaft seal to be subjected to frequent impact loads, making the sealing surface prone to microcracks. However, most current mud pump lifespan modeling methods are based on single-source information, such as fitting single-channel vibration or pressure data, static analysis of simulation results, and extrapolation of accelerated test data under fixed conditions.

[0004] These methods struggle to effectively adapt to the multi-factor coupled attenuation patterns of dredging vessels operating in complex environments, and the prediction models are prone to risks such as accuracy degradation, trend shifts, and early misjudgments. Therefore, a dredging pump life modeling method that can improve the accuracy and stability of dredging pump life prediction is needed. Summary of the Invention

[0005] Therefore, this invention provides a mud pump life modeling method based on multi-source residual collaborative learning to solve the problem of inaccurate mud pump life modeling in the prior art.

[0006] This invention provides a mud pump life modeling method based on multi-source residual collaborative learning, comprising: Obtain the output results of the simulation model simulating the operation of a mud pump; Collect multi-source data from the mud pump operating site; The residuals are calculated based on the multi-source data and the simulation model output to obtain the multi-source residuals; Based on the pre-set mud pump life model with multi-source residual correction, the actual mud pump life model is obtained.

[0007] In a preferred implementation: multi-source data includes flow field data and stress data; simulation model output includes flow field simulation results and stress simulation results; multi-source residuals include flow field residuals and stress residuals; residuals are calculated based on the multi-source data and simulation model outputs to obtain multi-source residuals, including: The residuals are calculated based on the flow field data and flow field simulation results to obtain the flow field residuals; The residuals are calculated based on the stress data and stress simulation results to obtain the stress residuals.

[0008] In a preferred implementation: flow field data includes pressure, gas holdup, and element concentration; flow field simulation results include CFD model simulation results; and flow field residuals include pressure residuals, gas holdup residuals, and element concentration residuals; stress data includes stress; stress simulation results include FEM model simulation results; and stress residuals include stress residuals.

[0009] In a preferred implementation: based on a pre-defined mud pump life model with multi-source residual correction, an actual mud pump life model is obtained, including: Multiple damage rate models based on multi-source residual correction presets; The preset mud pump life model is modified based on the modified multiple damage rate models to obtain the actual mud pump life model.

[0010] In a preferred implementation: multi-source residuals include flow field residuals and stress residuals; flow field residuals include pressure residuals, gas content residuals, and element concentration residuals; stress residuals include stress residuals; damage rate models include wear models, corrosion models, mechanical fatigue models, and cavitation models; and various damage rate models based on multi-source residual correction presets include: The wear model was modified based on the pressure residual and the gas content residual. The corrosion model was modified based on element concentration residuals and stress residuals; The mechanical fatigue model was modified based on stress residuals; The cavitation model was modified based on the gas content residual and pressure residual.

[0011] In a preferred implementation: the preset mud pump life model is a Weibull distribution curve fitted based on historical data; the preset mud pump life model is corrected based on various modified damage rate models to obtain the actual mud pump life model, including: By coupling and correcting multiple damage rate models, a synergistic damage rate model is obtained. Real-time cumulative damage is calculated based on a collaborative damage rate model. Based on the real-time cumulative damage, the Weibull parameters in the preset mud pump life model are corrected to obtain the actual mud pump life model.

[0012] In a preferred implementation: multiple modified damage rate models are coupled to obtain a cooperative damage rate model, including: Obtain the preset weights for each damage rate model. Obtain known damage information; Update the weights of each damage rate model based on the known damage information; Based on the updated weights, the modified damage rate models are weighted and superimposed to obtain a co-damage rate model.

[0013] This invention also provides a mud pump life modeling system based on multi-source residual collaborative learning, comprising: The simulation module is used to obtain the output results of the simulation model simulating the operation of a mud pump. The real-time monitoring module is used to collect multi-source data from the mud pump operating site; The residual analysis module is used to calculate residuals based on multi-source data and simulation model outputs to obtain multi-source residuals. The life modeling module is used to obtain the actual life model of the mud pump based on the preset mud pump life model with multi-source residual correction.

[0014] The present invention also provides an electronic device, comprising: Memory and processor; The memory is used to store the program, and the processor is used to execute the steps in any of the above-mentioned mud pump life modeling methods based on multi-source residual collaborative learning when the program is executed.

[0015] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement any of the steps in the above-described mud pump life modeling method based on multi-source residual collaborative learning.

[0016] The beneficial effects of adopting the above scheme are: This invention provides a mud pump life modeling method based on multi-source residual collaborative learning. First, the output of a simulation model simulating mud pump operation is obtained. Then, multi-source data from the mud pump's operating site are collected. Residuals are calculated based on the multi-source data and the simulation model output to obtain multi-source residuals. Finally, the preset mud pump life model is corrected based on the multi-source residuals to obtain the actual mud pump life model. This invention achieves accurate modeling and dynamic adaptation of mud pump life under complex operating conditions by constructing a closed-loop architecture of "simulation pre-running – measured feedback – residual correction – model evolution." The multi-source data fusion overcomes the limitations of monitoring single physical quantities, comprehensively covering the physical essence of multi-factor coupled damage in mud pumps. The residual-driven model correction mechanism quantitatively compares the theoretical output of the simulation model with measured data from actual operating conditions. By accurately identifying the deviation between model assumptions and the actual environment through residuals, the adaptability of the model to complex marine environments is significantly improved. Most importantly, this invention can continuously correct the mud pump life model based on real-time residuals, and has the ability to dynamically iterate and evolve. It is especially suitable for complex and ever-changing scenarios in the sea or waterways. The model can be updated without stopping the pump, providing core support for the intelligent operation and maintenance of dredging mud pumps that combines mechanistic depth and engineering practicality, and solving the problem of inaccurate mud pump life modeling in the existing technology. Attached Figure Description

[0017] Figure 1 The flowchart of the mud pump life modeling method based on multi-source residual collaborative learning provided by the present invention is shown below. Figure 2 for Figure 1 A detailed step diagram of step S104 is shown below; Figure 3 The system architecture diagram of the mud pump life modeling system based on multi-source residual collaborative learning provided by the present invention is shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Combination Figure 1 As shown, a specific embodiment of the present invention discloses a mud pump life modeling method based on multi-source residual collaborative learning, comprising: S101. Obtain the output results of the simulation model simulating the operation of the mud pump; S102. Collect multi-source data from the mud pump operating site; S103. Calculate the residuals based on the multi-source data and simulation model output results to obtain the multi-source residuals; S104. Based on the preset mud pump life model with multi-source residual correction, the actual mud pump life model is obtained.

[0020] In the above process, the simulation model output specifically represents the cumulative progressive damage of the mud pump material under set conditions, specifically the changes in parameters such as material properties and shape. Based on the simulation model output, a mud pump life model describing the changes in mud pump life can be summarized. It is understood that the specific representation of the simulation model and the mud pump life model can employ any existing technology. In this embodiment, the simulation model output can be a result simulated under pre-set conditions. If the ship's hardware computing capabilities allow, it can also be a result obtained through real-time simulation. The preset mud pump life model is obtained based on the simulation model output.

[0021] A more specific embodiment of the above process is as follows: 1. Multi-source signal fusion acquisition: Simultaneously acquire multiple sets of key variables representing the operating status of the mud pump, such as impeller wear clearance, sealing pressure, vibration acceleration, head, shaft power, and inlet / outlet pressure difference, and construct a multi-dimensional feature input matrix.

[0022] 2. Residual back-calculation and consistency correction mechanism: The residuals of on-site monitoring data and simulation model outputs such as CFD / FEM are calculated. Through weighted analysis, time drift alignment and residual trend learning, a multi-source information consistency correction algorithm is constructed to dynamically correct the model output.

[0023] 3. Construction and standardization of structural degradation indicators: A set of composite indicators that can characterize the degradation behavior of mud pumps are extracted, standardized and normalized to form the main control features for life modeling, and a degradation trajectory curve associated with the life distribution function is established.

[0024] 4. Dynamic lifetime modeling and recursive update algorithm: By introducing nonlinear Kalman filtering and recursive Bayesian estimation methods, the lifetime curve is continuously estimated and updated online, enabling adaptive adjustment of the model during long-term operation and avoiding the deviation of predicted values ​​over time.

[0025] 5. Interface with control system / SCADA platform: The model can be embedded in shipboard monitoring systems or back-end life management modules to achieve real-time status assessment, life curve projection, and early failure warning, providing data support for maintenance planning and operational rhythm optimization.

[0026] This embodiment has the following beneficial effects: Multi-source collaborative modeling improves prediction accuracy: By integrating simulation and measured data, the model bias is corrected through the residual back-calculation mechanism to ensure that the lifetime output has physical consistency; Dynamic updates adapt to changing operating conditions: It can adapt to non-constant operating conditions such as variable load, variable flow, and high disturbance, and achieve stability and sustainability of life estimation; It has closed-loop feedback capability: It supports adjusting model parameters according to the mid-term operating status to achieve continuous consistency between the prediction results and the actual degradation behavior; It can be embedded into existing system deployments: the algorithm is lightweight and the structure is universal, making it suitable for integration and application in SCADA, remote monitoring or digital twin platforms; Prediction accuracy is controllable: After residual regression optimization, the life prediction error of the model is controlled at a low level, which meets the accuracy requirements of engineering prediction and scheduling.

[0027] It is understandable that the simulation process of mud pump life decay can be basically divided into four stages: flow-induced damage simulation, structural response simulation, material degradation calculation, and life prediction (the specific implementation process of the four stages is existing technology and will not be elaborated on in this paper). Therefore, based on the above ideas, this invention provides some more optimized improvement schemes for the specific methods of residual calculation and model correction.

[0028] Specifically, in a preferred embodiment, the multi-source data includes flow field data and stress data; the simulation model output results include flow field simulation results and stress simulation results; and the multi-source residuals include flow field residuals and stress residuals.

[0029] Specifically, the flow field simulation results and stress simulation results correspond to the output results of the aforementioned "flow-induced damage simulation" and "structural response simulation," respectively. The flow field data and stress data are the measured data of the two simulation results. The flow field data can be any data related to the fluid flow field used for flow-induced damage simulation, such as inlet and outlet pressure, flow rate, mud concentration (solid content), particle size distribution, gas content, flow velocity, etc. It can be detected non-contactly / online by deploying pressure transmitters, electromagnetic flowmeters, online density meters / turbidity meters, laser particle size analyzers (non-contact), capacitive gas content sensors, etc., at locations such as pump inlet and outlet pipes, suction chamber, and discharge chamber. Stress-related data can be any data related to stress structure used for structural response simulation, such as vibration acceleration, acoustic emission signals, bearing / sealing cavity temperature, motor current / power, and oil metal particle concentration. It can be detected non-contactly / online by deploying piezoelectric accelerometers, broadband AE sensors, infrared thermal imagers / embedded thermocouples, current transformers, online oil spectrometers, etc., at locations such as pump housing / bearing housing, shaft system, oil tank, and motor output end.

[0030] The flow field residuals and stress residuals are the residuals corresponding to the two types of data mentioned above, respectively. The specific calculation method of the residuals can also be designed according to the actual situation. For example, the simplest method is to subtract the monitored multi-source data from the simulation results.

[0031] Based on this, step S103 above, which calculates the residuals according to the multi-source data and simulation model output results, yields the multi-source residuals, specifically including: The residuals are calculated based on the flow field data and flow field simulation results to obtain the flow field residuals; The residuals are calculated based on the stress data and stress simulation results to obtain the stress residuals.

[0032] This embodiment decouples the physical meaning of residual calculation by explicitly dividing multi-source data into "flow field" and "stress" categories and correspondingly decomposing the simulation model output into "flow field simulation results" and "stress simulation results." Specifically, flow field residuals focus on model deviations related to fluid-driven damage such as erosion, cavitation, and corrosion, while stress residuals address simulation deviations related to structural response-driven damage such as fatigue, mechanical wear, and stress concentration. This avoids the problem of mixing multi-physics noise with traditional single residuals. Furthermore, by independently calculating the two types of residuals in stages and correcting the corresponding stages of the model, the independent mechanisms of each physical process are preserved (such as the clear boundary between fluid mechanics and solid mechanics), while the residuals can accurately pinpoint specific simulation type errors (for example, flow field residuals often point to defects in the mud gas content model, while stress residuals often point to insufficient simplification of ship sway loads), making model correction more targeted. The improvements in this embodiment are particularly well-suited to the failure characteristics of dredging mud pumps, which are characterized by "multi-factor coupling but traceable origins." They not only reduce the complexity of residual analysis but also improve the reliability of damage rate formula correction through physical-level residual classification.

[0033] In a more specific embodiment, flow field data includes pressure, gas holdup, and element concentration; flow field simulation results include CFD model simulation results; and flow field residuals include pressure residuals, gas holdup residuals, and element concentration residuals. Stress data includes stress; stress simulation results include FEM model simulation results; and stress residuals include stress residuals.

[0034] CFD excels at simulating the flow characteristics of fluids (slurry), and flow field residuals are used to quantify the prediction bias of the CFD model against the actual flow field, correcting the flow field driving parameters for wear, cavitation, and corrosion. FEM excels at analyzing the stress-strain and fatigue response of structures. FEM can apply fluid loads based on CFD results, combined with mechanical loads, to calculate the stress and strain of the structure. Stress residuals are used to quantify the prediction bias of the FEM model against the actual structural response, correcting the structural driving parameters for fatigue and wear.

[0035] In this embodiment, the flow field data explicitly selected are "pressure, gas holdup, and element concentration," which correspond to the three core sub-processes in the CFD model: flow resistance characteristics, cavitation induction, and corrosive medium distribution, respectively. The pressure residual directly reflects the deviation between the actual flow channel blockage and the assumption of uniform flow field in the simulation; the gas holdup residual reveals the deviation between the actual gas entrainment and the gas evolution prediction of the cavitation model; and the element concentration residual (such as Cl...) - H + The concentration quantifies the misalignment between the chemical composition distribution of seawater / mud and the boundary conditions of corrosion simulation. Together, these three factors constitute a comprehensive residual diagnosis of flow-induced damage caused by "erosion-cavitation-corrosion".

[0036] Stress-related data focuses on "stress," which corresponds to the key outputs of structural dynamic response and load transfer in the FEM model. Stress residuals can not only capture the difference between actual stress concentration and simplified simulation load caused by ship swaying and shaft misalignment, but also infer mechanical damage driving deviations such as wear (e.g., local high-stress friction caused by uneven gap of the mouth ring) and fatigue (e.g., stress concentration in the impeller keyway) through stress distribution.

[0037] This embodiment's one-to-one correspondence between "parameters, process, and residuals" makes the residuals no longer abstract "data differences" but "fault clues" that can be directly traced back to specific physical mechanisms, greatly improving the interpretability and effectiveness of model correction, and is especially suitable for extreme working conditions where dredging pumps have strong coupling of multiple factors.

[0038] The above process can be based on a residual correction simulation model to obtain a more accurate mud pump life model. It is conceivable that after correcting the simulation model, a re-simulation is required, which is time-consuming and cumbersome in practice. Therefore, this invention also provides a more practical modeling method: Combination Figure 2 As shown, in a specific embodiment, step S104 above, obtaining the actual mud pump life model based on the preset mud pump life model with multi-source residual correction, specifically includes: S201. Multiple damage rate models based on multi-source residual correction presets; S202. Based on the modified multiple damage rate models, the preset mud pump life model is modified to obtain the actual mud pump life model.

[0039] This embodiment is essentially an improvement on the material degradation stage of the four stages described above: flow-induced damage simulation, structural response simulation, material degradation calculation, and lifetime prediction. Specifically, it uses a damage rate model based on multi-source residual correction to calculate material degradation, rather than the simulation process, significantly improving the practicality and timeliness of lifetime modeling. Specifically, this embodiment shifts the correction target from the complex simulation model to a lighter material degradation calculation module, directly calibrating the damage rate's adaptability to actual operating conditions through residuals, avoiding the high computational cost of repetitive simulations. Furthermore, since the damage rate model is a direct input for lifetime prediction, the corrected rate formula more accurately reflects the rate of material degradation at each moment under the current operating conditions. Combined with real-time monitored damage data, the lifetime distribution model can be rapidly iterated.

[0040] Specifically, based on the previous embodiments, the multi-source residuals include flow field residuals and stress residuals; the flow field residuals include pressure residuals, gas content residuals, and element concentration residuals; the stress residuals include stress residuals. In a new embodiment, the damage rate model includes a wear model, a corrosion model, a mechanical fatigue model, and a cavitation model. The above step S201, correcting the preset multiple damage rate models based on the multi-source residuals, specifically includes: The wear model was modified based on the pressure residual and the gas content residual. The corrosion model was modified based on element concentration residuals and stress residuals; The mechanical fatigue model was modified based on stress residuals; The cavitation model was modified based on the gas content residual and pressure residual.

[0041] This embodiment utilizes the combined effects of pressure and gas content residuals on the wear model to specifically correct the erosion coefficient and particle kinetic energy transfer efficiency. By jointly correcting the corrosion model using element concentration and stress residuals, it can simultaneously calibrate the activity of the corrosive medium and the stress-corrosion coupling coefficient. By separately correcting the mechanical fatigue model using stress residuals, it can directly correlate the deviation between the measured dynamic stress amplitude and the stress-fatigue curve. By synergistically correcting the cavitation model using gas content and pressure residuals, it can accurately capture the actual situation of cavitation generation and collapse in low-pressure areas. This one-to-one causal correction anchors the correction of each damage model to a clear physical cause. The residuals quantify the disturbance amplitude of actual working conditions on the mechanistic parameters, ultimately making the corrected damage rate model more closely reflect the attenuation nature of dredging pumps.

[0042] The present invention also provides a more specific embodiment to illustrate the above process more clearly: The wear model can be represented by the Bitter model: in, It represents the erosion wear volume per unit time and unit area, characterizing the wear rate; It represents the erosion coefficient of a material, which is related to the hardness and particle shape of the material of the flow-through component and reflects the material's resistance to erosion. This represents the particle impact velocity, which can be a preset value or obtained from CFD flow field simulation. The velocity index (usually 2-3) represents the sensitivity of the wear rate to particle velocity. Indicates the concentration of solid particles in the mud; This represents the impact angle function, describing the effect of the particle impact angle on the wall on wear.

[0043] The particle impact velocity and the concentration of solid particles in the mud can be corrected by the pressure residual and the gas content residual, respectively, to reflect the interference of actual flow resistance and gas content on particle motion. The corrected wear model is expressed as follows: in, Indicates pressure residual, Indicates the residual gas content. and These represent preset residual sensitivity coefficients, which can be obtained by setting preset values ​​or calibrating using historical fault data.

[0044] The corrosion model can be represented by an electrochemical (Faraday's law) model: in, It represents the mass loss due to corrosion per unit time and characterizes the corrosion rate. It represents the corrosion current density and reflects the electrochemical corrosion reaction rate; Indicates corrosion time; Indicates the molar mass of the material; Indicates the number of electrons transferred; is Faraday's constant.

[0045] The corrosion current density can be corrected using the element concentration residual and stress residual, respectively, to reflect the synergistic effect of elemental corrosion and stress corrosion. The corrected corrosion model is expressed as follows: in, Represents the residual of element concentration. Indicates stress residual, and These also represent the preset residual sensitivity coefficients.

[0046] Mechanical fatigue models can be represented by Miner's linear cumulative damage criterion: in, Indicates the first The actual number of cycles under the stress amplitude; The fatigue life under the i-th stress amplitude is calculated from the material's SN curve, specifically expressed as: in, The fatigue strength coefficient, The stress amplitude, This is the fatigue strength index.

[0047] The stress amplitude can be corrected by stress residuals to reflect the deviation between the actual dynamic stress and the simplified simulation load caused by ship swaying and shaft misalignment. The corrected mechanical fatigue model is expressed as follows: in, This represents the preset residual sensitivity coefficient.

[0048] Cavitation models can be represented by cavitation rate models: in, It represents the cavitation wear volume per unit time and per unit area, characterizing the cavitation rate; The cavitation erosion coefficient is related to the material toughness and the intensity of the cavitation collapse microjets. Indicates the density of the mud; This represents the impact velocity during cavitation collapse; This represents the cavitation volume fraction.

[0049] The cavitation volume fraction and the impact velocity during cavitation collapse can be corrected using the gas content residual and pressure residual, respectively, reflecting the influence of the actual gas content and the low-pressure zone pressure on cavitation. The corrected cavitation model is expressed as follows: in, and These also represent the preset residual sensitivity coefficients.

[0050] Further, in a preferred embodiment, the preset mud pump life model is a Weibull distribution curve fitted based on historical data. Based on this, step S202 above, which modifies the preset mud pump life model based on modified multiple damage rate models, yields the actual mud pump life model, specifically including: By coupling and correcting multiple damage rate models, a synergistic damage rate model is obtained. Real-time cumulative damage is calculated based on a collaborative damage rate model. Based on the real-time cumulative damage, the Weibull parameters in the preset mud pump life model are corrected to obtain the actual mud pump life model.

[0051] Specifically, the synergistic damage rate model can be obtained by weighted superposition of multiple damage rate models. Real-time cumulative damage can be obtained by integrating the synergistic damage rate model. The preset mud pump life model is specifically expressed as follows: in, Indicates time, This can be considered an injury. It is a natural constant. and All are Weibull parameters, where Indicates the scale parameter. Indicates shape parameters.

[0052] Arbitrary Weibull parameters can be corrected by accumulating damage in real time, for example: in, The corrected scale parameters. These are the corrected shape parameters. Indicates real-time cumulative damage. and All of these are preset sensitivity factors.

[0053] In this embodiment, the coupled and corrected wear, corrosion, fatigue, and cavitation rate models yield a synergistic damage rate model, quantifying the multi-factor interaction effect. Based on the synergistic damage rate, the real-time cumulative damage is calculated to invert and correct the Weibull distribution parameters. This not only retains the mature probabilistic statistical framework of the Weibull distribution in reliability engineering, but also realizes the dynamic evolution of the distribution parameters through the damage rate corrected by multi-source residuals and the real-time cumulative damage.

[0054] Further, in a preferred embodiment, the above step of coupling the modified multiple damage rate models to obtain a synergistic damage rate model specifically includes: Obtain the preset weights for each damage rate model. Obtain known damage information; Update the weights of each damage rate model based on the known damage information; Based on the updated weights, the modified damage rate models are weighted and superimposed to obtain a co-damage rate model.

[0055] The aforementioned known damage information includes both the damage to the mud pump and the known damage to the vessel carrying the mud pump. Unlike mud pumps in other fields, mud pumps on dredging vessels face short maintenance windows and difficulties in spare parts supply during offshore operations, making it difficult to perform maintenance at any time. Therefore, a "failure-based repair" strategy is typically adopted. For minor defects (such as seal leaks or slight wear), even if discovered, timely shutdown for repair may not be possible. Therefore, for mud pump life modeling on dredging vessels, known faults can be used to adjust the coupling ratio of multiple damage rate models to obtain a more accurate cooperative damage rate model, for example: If problems such as "abnormally high temperature in the sealing cavity" and "surge in Fe particle concentration in the oil" occur repeatedly recently, it indicates that mechanical fatigue and wear are the dominant failure modes at present. The weights of the fatigue model and the wear model can be significantly increased, while the weights of the corrosion and cavitation models can be reduced. If honeycomb-like pits appear on the impeller surface (observed by endoscopy), the weight of the cavitation model can be increased to accurately capture the temporary risk of cavitation aggravation induced by gas content fluctuations.

[0056] In this way, the collaborative damage rate model is no longer a static mathematical superposition, but a dynamic focus on the most likely and most easily deteriorated damage patterns, prioritizing the discovery of the damage types with the highest weights.

[0057] Combination Figure 3 As shown, the present invention also provides a mud pump life modeling system based on multi-source residual collaborative learning, comprising: Simulation module 310 is used to obtain the output results of the simulation model simulating the operation of the mud pump; The real-time monitoring module 320 is used to collect multi-source data at the mud pump working site; The residual analysis module 330 is used to calculate residuals based on multi-source data and simulation model output results to obtain multi-source residuals. The life modeling module 340 is used to obtain the actual mud pump life model based on the preset mud pump life model with multi-source residual correction.

[0058] It should be noted that the corresponding systems provided in the above embodiments are computer program products that can implement the technical solutions described in the above method embodiments. The specific implementation principles of the above modules or units can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0059] The present invention also provides an electronic device, comprising: Memory and processor; The memory is used to store the program, and the processor is used to execute the steps in any of the above-mentioned mud pump life modeling methods based on multi-source residual collaborative learning when the program is executed.

[0060] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement any of the steps in the above-described mud pump life modeling method based on multi-source residual collaborative learning.

[0061] This invention provides a mud pump life modeling method based on multi-source residual collaborative learning. First, the output of a simulation model simulating mud pump operation is obtained. Then, multi-source data from the mud pump's operating site are collected. Residuals are calculated based on the multi-source data and the simulation model output to obtain multi-source residuals. Finally, the preset mud pump life model is corrected based on the multi-source residuals to obtain the actual mud pump life model. This invention achieves accurate modeling and dynamic adaptation of mud pump life under complex operating conditions by constructing a closed-loop architecture of "simulation pre-running – measured feedback – residual correction – model evolution." The multi-source data fusion overcomes the limitations of monitoring single physical quantities, comprehensively covering the physical essence of multi-factor coupled damage in mud pumps. The residual-driven model correction mechanism quantitatively compares the theoretical output of the simulation model with measured data from actual operating conditions. By accurately identifying the deviation between model assumptions and the actual environment through residuals, the adaptability of the model to complex marine environments is significantly improved. Most importantly, this invention can continuously correct the mud pump life model based on real-time residuals, and has the ability to dynamically iterate and evolve. It is especially suitable for complex and ever-changing scenarios in the sea or waterways. The model can be updated without stopping the pump, providing core support for the intelligent operation and maintenance of dredging mud pumps that combines mechanistic depth and engineering practicality, and solving the problem of inaccurate mud pump life modeling in the existing technology.

[0062] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0063] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A mud pump life modeling method based on multi-source residual collaborative learning, characterized in that, include: Obtain the output results of the simulation model simulating the operation of a mud pump; Collect multi-source data from the mud pump operating site; The residuals are calculated based on the multi-source data and the simulation model output to obtain the multi-source residuals; Based on the pre-set mud pump life model with multi-source residual correction, the actual mud pump life model is obtained.

2. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 1, characterized in that, Multi-source data includes flow field data and stress data; simulation model outputs include flow field simulation results and stress simulation results. Multi-source residuals include flow field residuals and stress residuals; Residuals are calculated based on multi-source data and simulation model outputs to obtain multi-source residuals, including: The residuals are calculated based on the flow field data and flow field simulation results to obtain the flow field residuals; The residuals are calculated based on the stress data and stress simulation results to obtain the stress residuals.

3. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 2, characterized in that, Flow field data includes pressure, gas holdup, and element concentration; flow field simulation results include CFD model simulation results; and flow field residuals include pressure residuals, gas holdup residuals, and element concentration residuals. Stress data includes stress; stress simulation results include FEM model simulation results; and stress residuals include stress residuals.

4. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 1, characterized in that, Based on the pre-defined mud pump life model with multi-source residual correction, the actual mud pump life model is obtained, including: Multiple damage rate models based on multi-source residual correction presets; The preset mud pump life model is modified based on the modified multiple damage rate models to obtain the actual mud pump life model.

5. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 4, characterized in that, Multi-source residuals include flow field residuals and stress residuals; flow field residuals include pressure residuals, gas content residuals, and element concentration residuals; stress residuals include stress residuals. Damage rate models include wear models, corrosion models, mechanical fatigue models, and cavitation models; Multiple damage rate models based on multi-source residual correction presets include: The wear model was modified based on the pressure residual and the gas content residual. The corrosion model was modified based on element concentration residuals and stress residuals; The mechanical fatigue model was modified based on stress residuals; The cavitation model was modified based on the gas content residual and pressure residual.

6. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 4, characterized in that, The preset mud pump life model is a Weibull distribution curve fitted based on historical data; the preset mud pump life model is then corrected based on various modified damage rate models to obtain the actual mud pump life model, including: By coupling and correcting multiple damage rate models, a synergistic damage rate model is obtained. Real-time cumulative damage is calculated based on a collaborative damage rate model. Based on the real-time cumulative damage, the Weibull parameters in the preset mud pump life model are corrected to obtain the actual mud pump life model.

7. The mud pump life modeling method based on multi-source residual collaborative learning according to claim 6, characterized in that, By coupling and correcting multiple damage rate models, a synergistic damage rate model is obtained, including: Obtain the preset weights for each damage rate model. Obtain known damage information; Update the weights of each damage rate model based on the known damage information; Based on the updated weights, the modified damage rate models are weighted and superimposed to obtain a co-damage rate model.

8. A mud pump life modeling system based on multi-source residual collaborative learning, characterized in that, include: The simulation module is used to obtain the output results of the simulation model simulating the operation of a mud pump. The real-time monitoring module is used to collect multi-source data from the mud pump operating site; The residual analysis module is used to calculate residuals based on multi-source data and simulation model outputs to obtain multi-source residuals. The life modeling module is used to obtain the actual life model of the mud pump based on the preset mud pump life model with multi-source residual correction.

9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store the program, and the processor is used to execute the steps of any one of the mud pump life modeling methods based on multi-source residual collaborative learning in claims 1-7 when executing the program.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, are capable of implementing the steps in any of the mud pump life modeling methods based on multi-source residual collaborative learning as claimed in claims 1-7.

Images