Multimodal knowledge-driven production plant system multi-objective operating mode optimization method

Abstract

The application relates to the technical field of equipment operation and maintenance decision and optimization, and discloses a multi-modal knowledge driven production equipment system multi-objective operation mode optimization method, which comprises the following steps: constructing a production demand change prediction model based on periodical trend decomposition; constructing a multi-modal knowledge graph model; constructing a production equipment system operation quality evaluation index system and determining optimization targets; constructing a data driven equipment operation model based on an extended random flow network model; evaluating each optimization target of the production equipment system; constructing a multi-objective reward function based on a knowledge driven Pareto distance method; constructing an operation and maintenance management and mode optimization joint strategy optimization model; identifying and selecting key optimization targets and operation and maintenance technologies based on multi-modal production demand and production equipment operation data driven multi-modal knowledge graphs; and performing production equipment mode optimization and operation and maintenance management strategy joint optimization by using a deep deterministic policy gradient, so that operation and maintenance strategy dynamic optimization solving is realized.

Description

Technical Field

[0001] This invention relates to the field of equipment operation and maintenance decision-making and optimization technology, specifically to a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of production equipment systems. Background Technology

[0002] Currently, in Prognosis and Health Management (PHM), most methods for quantifying, modeling, and analyzing the health status of production equipment are based on the conformity view of equipment quality, which is the traditional definition of equipment status (also known as equipment quality) as the degree to which a set of inherent characteristics meet requirements.

[0003] As quality management has evolved from the production process quality management stage to the design process quality management stage, and then to the whole process quality management stage, the quality management level of various equipment has been continuously upgraded, and quality managers and researchers have gradually shifted from the concept of "conformity quality" to the concept of "suitability quality".

[0004] For continuously operating equipment with multiple objectives and high safety requirements (such as large and complex equipment in steel, chemical, pharmaceutical, and power systems), any unplanned downtime will affect the stable and safe operation of the entire system. Correspondingly, on-the-spot maintenance strategies such as Condition-Based Maintenance (CBM) and Predictive Maintenance (PdM) must be implemented within planned time constraints. Therefore, how to dynamically optimize production equipment operation and maintenance strategies under changing production demands has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of production equipment systems, in order to solve the problem of how to achieve dynamic optimization of production equipment operation and maintenance strategies under changing production demands.

[0006] This invention provides a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system, the method comprising:

[0007] A production demand change prediction model is constructed based on cyclical trend decomposition.

[0008] Construct a multimodal knowledge graph model for the production equipment system;

[0009] Construct a quality assessment index system for production equipment system operation, with the optimization objectives being basic equipment performance, remaining equipment lifespan, average maintenance interval, equipment operating time availability, equipment production capacity, equipment reliability, equipment operating safety, equipment output quality, equipment operation and maintenance costs and benefits, and equipment operating energy efficiency.

[0010] A data-driven device operation model is constructed based on the extended random stream network model.

[0011] Based on the extended random flow network model, the various optimization objectives of the production equipment system are evaluated;

[0012] Construct a multi-objective reward function based on a knowledge-driven method of distance between superior and inferior solutions;

[0013] Construct a joint strategy optimization model for operation and maintenance management and mode optimization;

[0014] Based on multimodal knowledge graphs, key optimization objectives and operation and maintenance technologies are identified and selected driven by multimodal production demand and production equipment operation data.

[0015] Based on deep reinforcement learning, a deep deterministic policy gradient is used to jointly optimize production equipment mode and operation and maintenance management strategies. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the application of a multi-modal knowledge-driven multi-objective operation mode optimization method for production equipment systems according to an embodiment of the present invention;

[0019] Figure 3 This is a flowchart illustrating the construction process of a multimodal knowledge graph model for a production equipment system according to an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of disturbance categories during the operation of a multi-state manufacturing system according to an embodiment of the present invention;

[0021] Figure 5This is a block diagram of an Extended Stochastic Flow Network (ESFN) model of the operation process of a production equipment system according to an embodiment of the present invention;

[0022] Figure 6 This is a basic framework diagram for multi-objective operation, maintenance and management of production equipment according to an embodiment of the present invention;

[0023] Figure 7 This is a schematic diagram of the ESFN operation model of a series-parallel structure production equipment system according to an embodiment of the present invention;

[0024] Figure 8 This is a flowchart illustrating the construction of a multimodal knowledge graph for equipment operation and maintenance management and mode optimization according to an embodiment of the present invention;

[0025] Figure 9 This is a framework diagram of the equipment system operation mode optimization and operation and maintenance management method according to an embodiment of the present invention;

[0026] Figure 10 This is a schematic diagram of the first part of the pseudocode of the DRL algorithm according to an embodiment of the present invention;

[0027] Figure 11 This is a schematic diagram of the second part of the pseudocode of the DRL algorithm according to an embodiment of the present invention;

[0028] Figure 12 This is a schematic diagram of an example and functional architecture of an open-pit mine mining and transportation equipment system according to an embodiment of the present invention;

[0029] Figure 13 This is a schematic diagram of MKG construction for operation and maintenance decision-making and mode optimization of open-pit mining and transportation equipment systems according to an embodiment of the present invention.

[0030] Figure 14 This is a heat map of MKG encoding results based on a graph neural network (GNN) according to an embodiment of the present invention;

[0031] Figure 15 This is a convergence graph of the training of a decision optimization algorithm model based on DRL according to an embodiment of the present invention;

[0032] Figure 16 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0034] As can be seen from existing research, the main problems with current research on equipment system operational status assessment technology are reflected in the following three aspects:

[0035] (1) The multimodal data, including text and language, generated during the operation of the equipment system is not fully utilized;

[0036] (2) The equipment status assessment indicators are relatively simple and still remain at the level of performance indicator compliance, lacking the assessment of the equipment's ability to complete production tasks, that is, the assessment method at the level of applicability.

[0037] (3) Lack of dynamic evaluation methods for system performance.

[0038] In terms of equipment system operation and maintenance decision-making and mode optimization technology, the existing research has the following two main problems:

[0039] (1) Operation and maintenance decision-making and mode optimization methods are not adaptable to changing production demands;

[0040] (2) There is an unreasonable maintenance decision based on "over-maintenance".

[0041] To address the aforementioned issues, this invention provides a multimodal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system. This method integrates multimodal operation data, considering various objectives such as production energy consumption, production efficiency, equipment failure rate, maintenance costs, product quality, and safety. Under varying demands, it optimizes the operation mode of the production equipment system by comprehensively applying technologies such as fault prediction and health management, intelligent operation and maintenance decision-making for manufacturing systems, and production scheduling and supply chain configuration, thereby achieving multi-objective adaptive mode optimization of the production equipment system.

[0042] According to an embodiment of the present invention, a method for optimizing the multi-objective operation mode of a production equipment system driven by multimodal knowledge is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] This embodiment provides a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system. The method steps are as follows: Figure 1 As shown. Figure 2 This is a detailed application flow of the multi-modal knowledge-driven multi-objective operation mode optimization method for production equipment systems according to an embodiment of the present invention. For example... Figure 1 , Figure 2 As shown, the process includes the following steps:

[0044] Step S101: Construct a production demand change prediction model based on cyclical trend decomposition.

[0045] In this embodiment of the invention, based on historical data and records, the Seasonal Trend Decomposition (STD) method is used to analyze, model, and predict the ever-changing production demand.

[0046] Step S102: Construct a multimodal knowledge graph model of the production equipment system.

[0047] In this embodiment of the invention, a multimodal knowledge graph (MKG) of the production equipment system is constructed, and multimodal operational data is introduced and integrated to achieve the selection of key optimization objectives and key operation and maintenance decision-making methods based on production needs and equipment status.

[0048] Step S103: Construct a quality assessment index system for the operation of production equipment systems, with the following optimization objectives: basic equipment performance, remaining equipment lifespan, average equipment maintenance interval, equipment operating time availability, equipment production capacity, equipment reliability, equipment operation safety, quality of finished products, equipment operation and maintenance costs and benefits, and equipment operation energy efficiency.

[0049] In this embodiment of the invention, as the basis for constructing a multi-objective operation model for a production equipment system, the performance indicators of the production equipment are extended to the applicability level in combination with production needs. Basic indicators such as finished product quality and delivery time are comprehensively applied, and the evaluation results are combined with indicators such as production energy consumption and maintenance costs to construct a multi-dimensional operation status evaluation system for the production equipment system.

[0050] Step S104: Based on the extended random stream network model, construct a data-driven device operation model.

[0051] In this embodiment of the invention, considering that the modeling of each operation optimization objective in step S103 depends on the real-time evaluation of the equipment's processing capacity, an Extended Stochastic Flow Network (ESFN) model is proposed to construct a data-driven equipment operation model to solve this problem.

[0052] Step S105: Based on the extended random flow network model, evaluate the various optimization objectives of the production equipment system.

[0053] In this embodiment of the invention, the equipment processing capacity calculated in step S104 is substituted into the model of each optimization objective in step S103 to evaluate various indicators such as the basic performance, production capacity, task reliability, operational safety, output product quality, cost and benefit, and operating energy consumption of the production equipment. The multi-objective operation mode optimization and dynamic operation and maintenance management method of the production equipment provide specific decision-making basis, and realize the construction of a data-driven equipment ESFN multi-objective operation model.

[0054] Step S106: Construct a multi-objective reward function based on the knowledge-driven superior-inferiority distance method.

[0055] In this embodiment of the invention, a multi-objective reward function based on the knowledge-driven weighted technique for order of preference by similarity to ideal solution (KW-TOPSIS) is constructed. While achieving multi-objective integration, the inference results of the multimodal knowledge graph on the importance of each optimization objective are applied to it in a weighted manner.

[0056] Step S107: Construct a joint strategy optimization model for operation and maintenance management and mode optimization.

[0057] In this embodiment of the invention, a joint strategy optimization model for operation and maintenance management and mode optimization is constructed, which includes operation mode control, material flow scheduling, energy flow scheduling and predictive maintenance.

[0058] Step S108: Based on the multimodal knowledge graph, identify and select key optimization targets and operation and maintenance technologies driven by multimodal production demand and production equipment operation data.

[0059] In this embodiment of the invention, based on the multimodal knowledge graph constructed in step S102, key optimization objectives and operation and maintenance technologies driven by multimodal production needs and equipment operation data are identified and selected.

[0060] Step S109: Based on deep reinforcement learning, deep deterministic policy gradients are used to jointly optimize production equipment mode and operation and maintenance management strategies.

[0061] In this embodiment of the invention, a Deep Deterministic Policy Gradient (DDPG) algorithm is developed based on Deep Reinforcement Learning (DRL) to achieve joint optimization of equipment mode optimization and operation and maintenance management strategies.

[0062] This embodiment provides a multi-modal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system. The process includes the following steps:

[0063] Step S201: Construct a production demand change prediction model based on cyclical trend decomposition.

[0064] Specifically, production demand is described through production task profiles. The production task profiles of the manufacturing system are described using four indicators: production time requirements, quality requirements, output requirements, and delivery integrity requirements (generally reflected in the loss due to stockouts of batch products; higher integrity requirements result in higher loss due to stockouts).

[0065] Considering the strong correlation between total output requirements and production time requirements, the production task profile is described by the following four relatively independent variables: j Batch product quality requirements for each production task (i.e., process capability requirements under Six Sigma theory), output per unit time requirements Production time requirements And batch product delivery integrity requirements, i.e., breach of contract penalty for unit shortage quantity. .

[0066] Therefore, a production task profile can be quantitatively described as a set of variables. , , , For time series ( (where is any one of the four variables), the general form of this model is:

[0067]

[0068] in, This represents the trend term of long-term changes in the time series. Periodic components are used to characterize the local periodic features of a time series. This is the accidental effect term. This represents the random error in the model estimation results.

[0069] Where, for any integer i Given a length of The observation time window The second-order difference observation sequence is as follows:

[0070]

[0071] Mostly follow a normal distribution .

[0072] in, The periodic component that characterizes the local periodicity of a time series can be approximated as a finite Fourier series.

[0073] Based on historical observations of the time series, the parameters of the above items are... Estimate the probability distribution of future terms of each variable in the production task profile to obtain the probability distribution.

[0074] Step S202: Construct a multimodal knowledge graph model of the production equipment system.

[0075] Specifically, the process of constructing a multimodal knowledge graph is as follows: Figure 3 As shown, the architecture of the multimodal knowledge graph is first designed. The multimodal knowledge graph for optimizing the device operation mode applied in this embodiment of the invention specifies production requirements, optimization goals and methods as entities, and the connections between them as relationships.

[0076] Specifically, production requirements include: (1) production capacity priority; (2) equipment service life guarantee priority; (3) production cost control priority; (4) operation and maintenance cost control priority; and (5) energy efficiency ratio priority.

[0077] The optimization objectives include basic equipment performance, remaining equipment lifespan, average maintenance interval, equipment uptime availability, equipment production capacity, equipment reliability, equipment operational safety, quality of finished product output, equipment operation and maintenance costs and benefits, and equipment operational energy efficiency.

[0078] The decision-making methods include equipment operation mode control optimization models, equipment material flow scheduling optimization models, energy flow scheduling optimization models, and equipment predictive maintenance optimization models.

[0079] Based on this, such as Figure 3 As shown in the "Knowledge Graph Architecture Design" section, it is necessary to acquire multimodal data returned from the field, including quantitative data (i.e., four types of operational data: equipment load fluctuation, performance degradation, energy flow intensity, and specific deviations in key product quality) and corpus data (including natural language descriptions of production demands and equipment operating status, i.e., raw corpus). Based on knowledge extraction technology, entity and relation identification are performed to complete entity and relation extraction, such as... Figure 3 As shown in the "Model Training and Knowledge Extraction" section. Therefore, based on historical cases, entities and relationships are extracted to construct a model such as... Figure 3The knowledge graph shown in the "Multimodal Knowledge Fusion and Reasoning Technology" section is then used. A Graph Neural Network (GNN) is introduced to complete the overall encoding of the multimodal knowledge graph, specifically the weight values ​​of the relationships between the various entities. Finally, knowledge reasoning is performed on the encoded knowledge graph to evaluate and rank the importance of 10 optimization objectives and 4 operational technologies. This enables the identification and selection of optimization objectives and operational technologies that meet production needs.

[0080] Step S203: Construct a quality assessment index system for the operation of production equipment systems, with the following optimization objectives: basic equipment performance, remaining equipment lifespan, average equipment maintenance interval, equipment operating time availability, equipment production capacity, equipment reliability, equipment operation safety, quality of finished product output, equipment operation and maintenance costs and benefits, and equipment operation energy efficiency.

[0081] Specifically, for the optimization objective (1) basic equipment performance: based on the random degradation process of the processing capacity of the production equipment and the interaction effect of work quality and equipment reliability generated during its operation, the basic performance of the production equipment is modeled and evaluated.

[0082] The specific modeling and evaluation process is as follows: A degradation model of the basic performance indicators of the equipment is constructed using the Gamma process. Specifically, in the [missing information - likely a specific timeframe or period] of the system's life cycle... j During the execution period of the production task, for the first production task in the system l The device can be described by a stochastic process model with values ​​in the range [0, 1], from a healthy state (1) to complete failure (0), that is:

[0083]

[0084] The model has a set of shape parameters:

[0085]

[0086] And the set of scale parameters:

[0087]

[0088] in, Indicates the first j The initial degradation during the execution period of each production task, the Gamma process. This represents the basic degradation of equipment performance over time, a process with shape parameters. With scale parameters .

[0089] For the optimization objective (2) remaining equipment life: based on the random degradation process of the basic performance of the production equipment system under the interaction effect of workpiece quality and equipment reliability during the operation of the production equipment, the remaining equipment life is predicted according to the principle of production equipment failure threshold.

[0090] The specific modeling and prediction process is as follows: Based on the basic performance degradation model:

[0091] Based on this, according to the current production task (whose serial number is denoted as ) j The minimum acceptable values ​​for the basic performance of the equipment must be clearly defined. This value can be considered as the failure threshold from the perspective of equipment suitability quality. Therefore, at the current moment... t Starting from the time reference point, the predicted remaining lifespan of the equipment can be recorded as:

[0092] in, Indicates an event The first occurrence of the event, i.e., the stochastic process Value reaches threshold Upon arrival. Correspondingly, the predicted remaining lifetime is: The expectation is estimated using the Monte Carlo method. Specifically, The numerical simulation can be repeated based on the probability model to estimate the result. The result obtained is recorded as follows: The remaining life prediction result of the equipment As shown in the formula above.

[0093] For the optimization objective (3) Mean Time To Repair (MTTR): Based on the aforementioned Remaining Life (RUL) prediction results, the MTTR of the equipment is modeled and evaluated according to the maintenance cycle set by the RUL quota triggering rules.

[0094] The specific modeling and evaluation process is as follows: Based on the numerical simulation process in the modeling process of the optimization objective (2), the remaining lifespan of the equipment starting from any time can be obtained. Therefore, based on this, equipment maintenance strategies can be set according to the Condition-Based Maintenance (CBM) model. Given the existence of RUL (Rating Limit Uptime) prediction, the common practice in CBM is to use RUL quota triggering rules, that is, to set a minimum acceptable RUL value. and in Predictions as low as Maintenance activities are triggered periodically. The failure threshold is considered in light of equipment load fluctuations, random degradation, and changes in production requirements. Change, function The temperature often does not decrease strictly linearly with time, but rather exhibits a certain degree of random fluctuation. Therefore, the time interval between maintenance trigger conditions is often not constant. Accordingly, during the simulation process... If considered as a stochastic process, the average maintenance interval can be evaluated based on the model of optimization objective (2) as follows:

[0095] in, Indicating the simulation process The k The second observation value, Indicates an event Total number of observations.

[0096] For optimization objective (4) Running Time Availability (RTA): based on the Mean Time To Repair (MTTR) prediction results, the average maintenance time is introduced. To model and evaluate RTA.

[0097] The specific modeling and evaluation process is as follows: According to the general definition of availability, equipment uptime availability refers to the proportion of equipment's uptime during its service life to its total service time. Therefore, the average maintenance time of the equipment can be calculated through historical maintenance records. Therefore, the assessment result of the equipment's uptime availability is as follows:

[0098] For optimization objective (5) equipment production capacity: Based on the performance margin model, evaluate the ability of the equipment system to produce products that meet the requirements of the task in terms of quantity and quality within a specified time, so as to realize the modeling and evaluation of equipment production capacity.

[0099] The specific modeling and evaluation process is as follows: This invention uses margin measurement to evaluate the production capacity of equipment, and the specific principle is as follows: Figure 4 As shown, where Figure 4 Figure (a) shows the required output of qualified products for the equipment's production task, as well as the maximum achievable output of qualified products for the equipment system during the task execution period. Figure 4 Option (b) provides corresponding capacity margin indicators as a measure and assessment result of equipment production capacity. Specifically, for the first [stage / phase] of the equipment system's life cycle... j The production task, as shown in the production task profile, has a required output of qualified products of the following specifications: Therefore, the equipment should meet the quality requirements of the finished product. Under the condition of yield of qualified products The equipment production capacity index, characterized by capacity margin, can then be denoted as:

[0100]

[0101] in, The evaluation relies on the constructed ESFN model. It can be seen that the yield of qualified products... With equipment production capacity Directly related, therefore , The value of is also affected by the decision variable The impact of this. Therefore, the production capacity index given in the above formula can be written as:

[0102] For optimization objective (6) equipment reliability: from the perspective of equipment functional suitability quality, it is defined as task reliability, and its functional suitability is evaluated based on capacity margin, thereby assessing the task reliability index of the production equipment system.

[0103] The specific modeling and evaluation process is as follows: Production equipment reliability is defined as the ability to produce batches of products that meet demand in both quantity and quality under specified conditions and within a specified time. Therefore, the condition of "production equipment reliability" is equivalent to the capacity margin index constructed above. >0. Specifically, to assess the reliability of production equipment tasks, the production task profile should first be predicted using the proposed STD model. The probability distributions are denoted as follows: Then the device in the first j Within the production task cycle, i.e., 0≤ t ≤ T ( j The reliability index within the specified time period is:

[0104]

[0105] in, .

[0106] For the optimization objective (7) equipment operation safety: considering the uncertainties in the equipment operation process, based on the traditional FMEA (Failure Mode and Effects Analysis) qualitative modeling method, a probability distribution model of the severity, frequency and undetectability of failure consequences is introduced, and a probability-based risk priority number (PRPN) index model is constructed to realize the quantitative evaluation of safety risks considering uncertainties, and to evaluate the operation safety of the equipment based on this.

[0107] The specific modeling and evaluation process is as follows: First, assess the severity (S) of the consequences of equipment failure.

[0108] This section presents different methods for assessing the severity of failure consequences based on various application scenarios. In engineering applications, decision-makers should select the appropriate method to assess the expected severity of failure consequences according to the specific circumstances.

[0109] (a) Situation where equipment failure incident data is available: After obtaining the severity data of the failure consequences, sort them from smallest to largest as follows: x 1, x 2…… x n}, then the observed data x i Let the theoretically possible range of values ​​for the data be ( ). a , b The cumulative distribution function (CDF) for severity should be:

[0110]

[0111] in, .

[0112] Based on uncertainty theory, by transforming the solution algorithm on uncertainty measures to probability measures, the cumulative probability distribution of severity can be solved. c ( x The numerical approximation solution is obtained, and then the expected severity of the consequences of equipment failure is obtained:

[0113]

[0114] (b) Cases where no equipment failure incident data is available. In this case, expert evaluation information should be used to assess and estimate the severity of the consequences of the equipment failure. Specifically, suppose the first [item name] in the system... l Severity of consequences caused by equipment failure There are respectively n Possible values ​​{ x 1, x 2…… x n (Sorted from smallest to largest), and has m If several experts participate in the severity assessment of the consequences of equipment failure, then for the first... l The Delphi method for assessing the severity of the consequences of equipment failure can be carried out as follows: (1) Obtain the expert reliability rating matrix for the severity of the failure consequences: Among them, reliability score Indicates the first i Experts on the incident (2) For the expert reliability rating matrix, calculate the mean reliability rating of each expert column by column: Standard deviation (3) Conduct a consistency test of expert reliability scores: given the maximum error tolerance If and only if Accept the expert evaluation results immediately; otherwise, repeat steps (1) and (2); (4) After passing the consistency test, construct the subjective probability distribution of the severity of the equipment failure consequences according to the following formula:

[0115]

[0116] Solve according to the above distribution. Expectations: This refers to the assessment result of the severity of the consequences of equipment failure.

[0117] Next, assess the failure frequency (O) during equipment operation.

[0118] The following assessment of the failure frequency (O) during equipment operation is based on a proportional hazards regression model. For the... l This device, here we assume its initial degradation value. Degradation rate (using scale parameters) They approximately follow a normal distribution. , , Then the probability density function of the time of functional failure is:

[0119]

[0120] Furthermore, the device survival function corresponding to the implementation of this function can be expressed as:

[0121]

[0122] Accordingly, the failure rate function can be expressed as: Building upon this, the continuous degradation phenomena of non-critical components within the equipment (such as guide rails, lead screws, and fixtures) are further expressed as covariates (including the degradation process vector of non-critical components). and the corresponding regression coefficients By integrating it into the failure rate model, we can obtain the first... l Comprehensive functional failure rate model of safety-critical equipment during operation At the same time, the failure probability model of the equipment during operation was obtained:

[0123] The undetectability (D) of the equipment failure modes is then assessed. The assessment of the detectability of different equipment failure modes relies on the experience of on-site maintenance personnel in the production workshop. Therefore, for failure modes... l Let its undetectability be denoted as For the undetectability index of equipment failure modes, the failure consequence severity evaluation method mentioned above can be applied to assess the expected outcome. .

[0124] Therefore, the safety of equipment operation can be assessed based on risk indicators as follows: Based on the RPN model of equipment operation risk, the equipment... l Operational risk is defined as the probability of failure during operation. Severity of failure consequences The product of . Therefore, for the common M For a production equipment system of a certain machine, its operational risk can be defined as:

[0125] Since operational risk values ​​are generally expected to be small, while in engineering practice safety indicators should generally be described as having large characteristics.

[0126] Specifically, one feasible approach is to: give an operational risk baseline value. And based on this, a dimensionless model is used. Establish safety assessment indicators for equipment operation.

[0127] For the optimization objective (8) equipment output quality: introduce the quality reliability (QR) propagation chain model to model and evaluate the equipment output quality.

[0128] The specific process is as follows: by process i The first one processed in k Taking one key quality characteristic (KQC) as the research object, and establishing a process response model that considers the parameter design prioritization principle, then the deviation model of this KQC is... It can be represented as:

[0129]

[0130] in, It is the baseline constant. It is the vector of machine degradation factors. This is the equipment operating noise vector. Accordingly, the deviation index of this KQC can be defined as follows:

[0131] For any KQC k According to the upper limit USL and lower limit LSL of the specification and the process capability index requirements. A deviation threshold can be obtained. The final quality of a product depends on the processing results of all key quality characteristics, therefore the equipment... i The output product qualification rate can be expressed as:

[0132]

[0133] For optimization objective (9) equipment operation and maintenance costs and benefits: Consider equipment maintenance costs, production task delay and default losses, overtime losses of maintenance activities and quality problems in the finished product use stage, and evaluate the equipment system operation and maintenance costs and benefits.

[0134] Specifically, this includes: (1) the basic revenue generated by a given unit output of qualified products is (Note: The initial revenue from completing a production task is determined by a variety of factors, including the product's market price, raw material costs, labor costs, equipment operating costs, and logistics costs.

[0135]

[0136] (2) Equipment maintenance costs. Production equipment undergoes inspection and maintenance during production task intervals. The maintenance options for each piece of equipment include minimal repair, imperfect repair, corrective repair, and replacement. The time spent and unit time cost for these four activities are recorded as follows: , , , as well as , , , Then the first j The equipment maintenance cost after the production task execution period can be recorded as:

[0137]

[0138] in, , , , , , , , These are maintenance decision variables, representing the corresponding maintenance decisions (i.e., minimum maintenance (MI), imperfect maintenance (IM), corrective maintenance (IM), replacement (RP)) applied to production equipment. (1) No (0) l superior.

[0139] (3) Production task delay and default loss. Considering that production equipment may fail to complete production tasks smoothly in the event of unexpected failures or sudden degradation, the delay and default loss must be considered in the operation and maintenance decision-making. That is, the default loss caused by the system's inability to deliver products on time and in sufficient quantity under the current product quality requirements and production time requirements. As can be seen from the relevant model of optimization objective (2), in the firstj At the end of the production task execution period, the expected shortage of qualified products is:

[0140]

[0141] in, For equipment l The equivalent qualified product output at the system output can be calculated using the ESFN model. Therefore, considering the first... j Loss per unit of stockout given in the production task profile (Approximately taken as the expected value estimated by the STD model) The assessment of the default loss due to production task delay is as follows:

[0142]

[0143] (4) Losses due to overtime of maintenance activities. For production and manufacturing equipment systems in process industries, maintenance activities are generally limited to the time constraints of specific production task intervals (denoted as...). T cons Within this period. If the maintenance activity exceeds this constraint, it will encroach on the execution time of the next production task, leading to an increase in the penalty for delays in the next production task. Considering that maintenance activities for different equipment can generally be carried out simultaneously, therefore, in the first... j The total time spent on maintenance and repair activities after the production task execution period is:

[0144]

[0145] Accordingly, the expected timeout is:

[0146]

[0147] The increment of the default loss due to the delay in the next round of production tasks caused by this overtime is the overtime loss for this round, denoted as:

[0148]

[0149] (5) Losses due to quality issues during the finished product usage phase. Considering that defective materials are unavoidable in the finished products produced by the production equipment system, and that these defects are exacerbated into quality problems during the usage phase, specifically manifested as higher-than-expected failures within the warranty period, the early failure rate of finished products manufactured by the production equipment can be assessed using the Cox proportional hazards model, taking into account the patterns and categories of material defects:

[0150]

[0151] in, This represents the baseline failure rate (i.e., the failure rate of a defect-free product under normal operating conditions). This represents the rate of missed defective material inspection in processes 1 to M of the production equipment system. express The corresponding proportional risk regression coefficient.

[0152] Record the warranty period as Therefore, the reliability of the finished product during the warranty period is:

[0153] Therefore, considering that the total output of the finished product is Therefore, the loss due to quality problems during the use of the finished product is:

[0154]

[0155] in, This represents the ratio of losses due to stockouts of the same quantity of products to economic losses caused by quality issues during the usage phase.

[0156] Taking into account the above four types of costs and losses, it can be seen that the equipment system can achieve the following by executing the first... j Expected net revenue from each production round:

[0157] This is the indicator of the operating and maintenance revenue (cost) of production equipment.

[0158] For the optimization objective (10) equipment operation energy efficiency: considering the four stages of production equipment system in process industry, namely start-up, preheating, operation and shutdown, the energy efficiency ratio index is applied to model and evaluate the equipment operation energy efficiency index.

[0159] The specific evaluation process is as follows: Considering that, apart from the working state, the start-up, preheating, and shutdown states all have the following characteristics: short duration, low total energy consumption, and are independent of equipment performance status, the energy consumption in these three states can be considered as the same constant for any production task. Energy consumption during operation is closely related to the evolution of equipment performance during operation. Existing energy consumption modeling methods for production equipment typically use the following model: This describes the change in energy consumption rate of a polymorphic degrading system during performance degradation. Indicates equipment l The baseline energy consumption (this baseline is a decision variable and is controlled by the decision-maker in the optimization of the operating mode); This represents the state-energy consumption rate impact factor model of the equipment. In the equipment basic performance degradation model... Based on the impact factor It can be defined as a power function model: .

[0160] Among them, constant coefficients This can be approximated using regression calculations. Therefore, the equipment... l In the j Cumulative energy consumption during the production task execution period:

[0161]

[0162] And the energy efficiency ratio of the equipment system (the energy consumed to obtain a unit of net revenue):

[0163] In actual equipment operation and maintenance decision-making in engineering projects, it can be Incorporating energy efficiency indicators into operation and maintenance decision-making to assess the energy flow carried by equipment. To allocate, control, and optimize.

[0164] Step S204: Based on the extended random stream network model, construct a data-driven device operation model.

[0165] Specifically, the evaluation process is as follows: The ESFN model divides the material flow in the equipment system into four states: "Perfect" (O), "Scrap" (S), "Unqualified" (U), and "Passes Quality Inspection but Has Defects" (F). Production equipment is categorized into two types: those that support rework and those that do not. In equipment that does not support rework, quality state U is equivalent to S. In equipment that supports rework, materials with quality state U have one and only one chance to be reworked. In the quality inspection stage of the equipment system, the direct basis for determining whether a material can pass quality inspection is the product KQC value. For the first... l The KQC term, given a baseline tolerance, is... Then, in the production task profile requirements Under these conditions, the pass rate for a single quality inspection is:

[0166]

[0167] Based on this, it can be concluded that: for equipment l In this case, considering the possibility of multiple types of material inputs, let the upstream adjacent process equipment set be denoted as . ,but:

[0168] (1) If the equipment supports the rework of non-conforming materials, its ESFN model can be described as follows:

[0169] (2) If the equipment does not support rework of non-conforming materials, its ESFN model can be described as:

[0170] in, Indicates equipmentl The corresponding defective part omission rate in the quality inspection process. Based on the above formulas, as follows... Figure 5 From (b), it can be seen that if the equipment l If rework of non-conforming materials is supported, the overall processing pass rate will be:

[0171]

[0172] Conversely, if it does not support the rework of defective materials, then its overall processing pass rate should be:

[0173]

[0174] Therefore, recording equipment l The complete set of downstream process equipment is Therefore, from the above content, it can be concluded that: from the perspective of the final output of the equipment system, the equipment... l Production task requirements Under these conditions, its equivalent processing capacity is:

[0175]

[0176] in, .

[0177] Therefore, for a system consisting of M production machines arranged in series or multiple series configurations, the maximum achievable processing capacity per unit time is:

[0178] .

[0179] Step S205: Based on the extended random flow network model, evaluate the various optimization objectives of the production equipment system.

[0180] Specifically, the calculation in the previous step By substituting the models of the optimization objectives in step S203, we can evaluate various indicators of production equipment, including basic performance, production capacity, task reliability, operational safety, output quality, cost and benefit, and operational energy consumption. This provides a concrete decision-making basis for optimizing the multi-objective operation mode and dynamic operation and maintenance management methods of production equipment. Thus, this chapter realizes the construction of a data-driven multi-objective operation model for equipment ESFN.

[0181] Step S206: Construct a multi-objective reward function based on the knowledge-driven superior-inferiority distance method.

[0182] Specifically, step S206 includes:

[0183] Step S2061: Normalize the optimization target of the equipment operation mode.

[0184] In this embodiment of the invention, the operation and maintenance optimization objectives, which have different dimensions and values, are first subjected to dimensionless and normalized processing.

[0185] Since the dimensions of the optimization objectives for different operating modes are different, they need to be normalized first; otherwise, assigning weights to the importance of each optimization objective will be meaningless. Specifically, for the optimization objectives... x In this regard, the normalization method is:

[0186]

[0187] in, For the normalization operator, , They represent the optimization objectives respectively. x The theoretically achievable maximum and minimum values. Following this method, the optimization objective of the equipment operation mode (i.e., the operation quality evaluation index) is normalized as:

[0188]

[0189] Step S2062: Extract nodes and association models from the multimodal knowledge graph, and extract the strength weight values ​​of the associations between various entities.

[0190] In this embodiment of the invention, nodes and association models are extracted from the multimodal knowledge graph in step S202 and imported into this step. At the same time, based on the aforementioned GNN encoding results, the strength weight values ​​of the associations between entities are extracted and imported as attributes of the association edges.

[0191] Step S2063: Determine the activation nodes of the knowledge graph and calculate the weighted optimization target vector.

[0192] In this embodiment of the invention, the activation node of the knowledge graph is identified based on the current equipment type, operating status, and production demand scenario. Starting from the activation node, the importance weight of the optimization target is derived. Based on the derivation result, the target with a weight higher than a certain threshold is adopted as a key target in the operation and maintenance decision-making, and its weight coefficient is applied to the weighted integration of the objective function.

[0193] Specifically, step S2063 includes:

[0194] Step S20631: Determine the activation node of the knowledge graph based on the current equipment type, operating status, and production demand node category.

[0195] In this embodiment of the invention, the "activation node" of the knowledge graph is determined based on the current specific application scenario, namely the specific category of the device type, operating status, and production demand node.

[0196] Step S20632: Starting from the activated node, traverse the predefined associated paths and calculate the importance of the optimization target.

[0197] In this embodiment of the invention, the process starts from the activated node and traverses along a predefined associated path. The weights on the path are multiplied or calculated using a weighted summation aggregation function. The importance of the "optimization target" node is evaluated as the sum of the cumulative weights of all paths that can reach it and originate from the activated node, i.e.:

[0198] in, For the first i One optimization objective, For the first i The importance of each optimization objective For input ( input To the optimization objective Path weights.

[0199] Step S20633: The importance of the optimization objective is applied to the objective function for weighted integration to obtain the weighted optimization objective vector.

[0200] In this embodiment of the invention, the weight set After evaluation and normalization, it can be applied to the selection and weighted integration of optimization objectives. Specifically, the weight coefficients of each optimization objective are denoted as a vector:

[0201]

[0202] For optimization objectives that were not selected, the weights were reassigned. =0, which means we get the weighted optimization objective vector:

[0203] The weighted TOPSIS approach integrates multiple optimization objectives, transforming the multi-objective equipment operation mode optimization problem into a single-objective optimization problem. It then selects appropriate operation and maintenance management methods to optimize the equipment operation mode. The specific principles are as follows: Figure 6 As shown. In particular, in actual production, there are often situations where the knowledge graph only identifies a single optimization objective. In such cases, production and operations decision-makers can directly skip the process of importing objective weight values ​​and transform the pattern optimization problem into a single-objective problem.

[0204] Step S2064: Calculate the Euclidean distance between the weighted optimization objective vector and the optimal and worst solutions in the objective space, and construct a multi-objective reward function.

[0205] In this embodiment of the invention, the following formula is first calculated: The distance to the optimal and worst solutions in the objective space. In the operational mode optimization problem, since each optimization objective has the characteristic of maximizing the objective value, it can be known that... The Euclidean distances to the optimal and worst solutions in the target space are respectively:

[0206]

[0207] in, To calculate the Euclidean distance of the optimal solution of the target vector in the target space using weighted optimization. To calculate the Euclidean distance of the worst solution of the objective vector in the objective space using weighted optimization. To optimize the target vector using weighted methods, The optimal solution is the zero vector. This is the worst solution.

[0208] Because the TOPSIS method assumes that under a certain optimization scheme, The farther the solution is from the worst-case scenario and the closer it is to the best-case scenario in the target space, the better the solution. Therefore, based on the calculations above... The distance to the optimal and worst solutions can be used to construct a multi-objective reward function for optimizing the device system's operating mode, as shown below:

[0209]

[0210] By maximizing this reward function as the optimization objective, an algorithm can be designed to achieve multi-objective operation and maintenance management and mode optimization for production equipment.

[0211] Step S207: Construct a joint strategy optimization model for operation and maintenance management and mode optimization.

[0212] Specifically, focusing on the key operational objectives of the equipment, and combining the industrial big data platform, the equipment operation observation data is applied to select appropriate mode optimization and operation and maintenance management methods, construct a decision optimization model, and realize the joint optimization of equipment operation mode control, load scheduling, energy allocation and predictive maintenance. First, as can be seen from the aforementioned equipment operation quality assessment index system, the joint strategy optimization framework of equipment operation and maintenance management and mode optimization mainly includes the following methods: (1) equipment operation mode control; (2) equipment system material flow scheduling; (3) equipment operation energy flow scheduling; (4) equipment predictive maintenance.

[0213] The following steps will model the optimization problems described above:

[0214] Decision Model (1) Equipment Operation Mode Control Optimization Model.

[0215] The specific modeling process is as follows: For multi-stage manufacturing processes, production workshops often equip each stage with multiple parallel devices as redundancy backups to ensure adaptability to constantly changing production demands. In this case, the ESFN operation model of the equipment system can be extended as follows: Figure 6 As shown in the diagram. When product demand increases rapidly in a short period, production decision-makers need to promptly activate backups to increase output; conversely, they need to switch backup devices from running mode to standby or offline mode in a timely manner. Specifically, for the first... l The first step of the process k Parallel equipment (numbered as) l . k Producers should rationally control the operating mode (online operation, standby reserve, or offline) of each piece of equipment based on production needs and equipment status, i.e., mode variables:

[0216] To effectively ensure the basic performance of the equipment Production capacity Mission reliability Operational safety Net production income expectations Record the first l The total number of equipment in the process is Then its equipment operation mode can be described as a matrix of decision variables:

[0217] The matrix satisfies the following constraints: .

[0218] Decision model (2) is an equipment material flow scheduling optimization model.

[0219] The specific modeling process is as follows: To reasonably control the operating load and performance degradation of multiple parallel devices within each process, and to ensure the successful completion of production tasks, the producer needs to rationally schedule material flow and control equipment operating load. Specifically, when scheduling operating load, the producer needs to rationally allocate the execution of the first... l The first step of the process l The load proportion allocated to each parallel device in the subsystem. For example... Figure 7 As shown, in runtime load scheduling, the decision variable directly controlled by the producer is the variable allocated to the first... l In the process of the first j Equipment (abbreviated as No.) l . j Material flow ratio of equipment For the first l In terms of process, it can be represented as a decision variable vector:

[0220]

[0221] Given the conditions: (1) the material flow ratio should meet the normalization requirements; (2) the material flow ratio allocated to equipment in standby or offline state should be 0, it can be seen that the constraints that the equipment operating load ratio vector should meet are:

[0222]

[0223] in, M This is a very large positive number. Since the load directly affects the rate of degradation of equipment performance, and thus affects the frequency of equipment failure, production capacity and task completion ability, when decision-makers are concerned about the basic performance of equipment, production capacity, task reliability, operational safety and expected net production benefits, they should consider the dynamic optimization and adjustment of equipment operation load scheduling during operation and maintenance management.

[0224] The decision model (3) is an energy flow scheduling optimization model.

[0225] The specific modeling process is as follows: Based on operation mode control and load scheduling, in order to optimize the energy efficiency indicators of the equipment system and ensure the stability of its output product quality, production decision-makers often need to synchronously schedule the energy flow carried by parallel equipment in each process. The energy flow scheduling model is similar to the operation load scheduling model; production decision-makers should schedule the proportion of energy flow carried by each parallel device in each process (for the first...) l . j Equipment is recorded as The allocation is performed, and its decision variable vector is:

[0226]

[0227] Similar to the conditional constraint model of load scheduling, the scheduling of equipment operating energy flow should satisfy the following constraints:

[0228]

[0229] As can be seen from the derivation in step S203, the energy flow and its fluctuations during equipment operation will affect the stability of product quality. Energy efficiency of equipment system operation This has a significant impact. Therefore, when these two optimization objectives are given significant importance, production decision-makers should incorporate the scheduling of equipment operating energy flow into the comprehensive decision-making model for operation and maintenance management and mode optimization.

[0230] Decision model (4) is a predictive maintenance optimization model for equipment.

[0231] The specific modeling process is as follows: For a given piece of equipment, the predictive maintenance options available during the production task interval include Minimal Maintenance (MI), Imperfect Maintenance (IM), Repair Maintenance (IM), and Replacement Maintenance (RP). j During the production task execution interval, for the first l . j Regarding the equipment, they are respectively , , , Therefore, for the first l For the subsystem of the process, the predictive maintenance scheme for equipment can be expressed as the following decision variable matrix:

[0232] The constraints it satisfies are:

[0233] .

[0234] As can be seen from the optimization model above, the overall equipment operation and maintenance management and mode optimization methods encompass four categories of operation and maintenance decision-making methods: equipment operation mode control, load scheduling, energy flow scheduling, and predictive maintenance decision-making. When implementing optimization decisions, these four methods share the same optimization objective:

[0235]

[0236] Therefore, joint optimization is convenient and feasible. Based on the aforementioned modeling analysis of each operational optimization objective, decision-makers can apply knowledge graphs to identify and evaluate key optimization objectives in actual production, thus selecting from the four methods mentioned above.

[0237] Step S208: Based on the multimodal knowledge graph, identify and select key optimization targets and operation and maintenance technologies driven by multimodal production demand and production equipment operation data.

[0238] based on Figure 8 Construct a multimodal knowledge graph to identify and select key optimization objectives and operation and maintenance technologies driven by multimodal production needs and equipment operation data.

[0239] Specifically, step S208 includes:

[0240] Step S2081: Based on entity information and relation list, form entity information, relation list and triple list.

[0241] In this embodiment of the invention, firstly, entity lists, relationship lists, and triple lists are formed using entity information and relationship information from a historical case database, such as... Figure 8 As shown in (a).

[0242] Step S2082: Using the triplet list and device operation data as input, encode the entities and relations to obtain entity encoding vectors and relation encoding vectors.

[0243] In this embodiment of the invention, with Figure 8 (a) The triplet list and the aforementioned four types of equipment operation data (equipment load fluctuation, performance degradation, energy flow intensity, and finished product KQC deviation) are used as input. For each entity and relation, vectorized encoding is performed, such as... Figure 8 As shown in (b). In the entity and relation encoding here, except for the encoding vector of the device entity which comes from the running data, the other entity and relation encodings all come from the text corpus information in the historical case library.

[0244] Step S2083: Calculate entity similarity and relationship similarity using similarity calculation, assign weights to the relationships between entities, and form a weighted graph model.

[0245] In embodiments of the present invention, such as Figure 8 As shown in (c), similarity measurement methods are used to match the similarity between three types of entities: equipment, production demand and optimization target. The relationships between entities are weighted and a weighted graph model is formed.

[0246] Step S2084: Based on the weighted graph model, the importance of the optimization objectives and operation and maintenance technologies is ranked, and the optimization objectives and operation and maintenance technologies are selected.

[0247] In this embodiment of the invention, graph neural networks and other technologies are used to obtain the importance ranking and evaluation of optimization objectives and operation and maintenance technologies, and to achieve the identification and selection of optimization objectives and operation and maintenance technologies based on importance. Figure 8 As shown in (d).

[0248] Step S209: Based on deep reinforcement learning, deep deterministic policy gradients are used to jointly optimize production equipment mode and operation and maintenance management strategies.

[0249] Specifically, step S209 includes:

[0250] Step S2091: Construct a Markov decision model for multi-objective operation and maintenance management and mode optimization of production equipment, and determine the state space, action space and reward function of the decision-making process.

[0251] In embodiments of the present invention, a construction is performed as follows: Figure 9The Markov Decision Process (MDP) model for multi-objective operation and maintenance management and mode optimization of production equipment shown clearly defines the state space, action space, and reward function of the decision-making process. The action space consists of the decision results selected from the four types of methods constructed in step S207 and processed through step S208. The reward function is constructed using the TOPSIS method in step S206, while the state space is the embedded state tensors of the device's operating process. These tensors are obtained by encoding the big data of the device's operating process through the encoder. The big data of the operating process includes three categories: observation data of fluctuations in output product quality characteristics, observation data of device performance degradation, and observation data of fluctuations in device energy flow. The encoder can adopt a one-dimensional convolutional neural network model.

[0252] Step S2092: Design the Actor-Critic network model as a deep reinforcement learning agent.

[0253] In this embodiment of the invention, a deep reinforcement learning agent network model is designed, employing an actor-critic network model as the DRL agent, and using Long Short Term Memory (LSTM) and Multilayer Perceptron (MLP) networks as the architectures of the actor network and critic network, respectively. Figure 9 As shown in the diagram, in this intelligent agent, the actor network is equivalent to a decoder, taking the device operating state tensor as input and outputting actions, i.e., the decision results of operation and maintenance management and mode optimization. The critic network takes both the operating state tensor and the decision results as input and outputs a reward function, which is used to fit the functional relationship between the state, action, and reward.

[0254] Step S2093: The deep deterministic measurement gradient algorithm is used to train the agent network and output the production equipment operation mode optimization and operation and maintenance management strategy.

[0255] In this embodiment of the invention, a deep reinforcement learning algorithm is designed. To ensure the adaptability and convergence of the algorithm during operation, a Deep Deterministic Policy Gradient (DDPG) algorithm is adopted. Combined with equipment system operation simulation, the agent network is trained online so that the critic network can correctly fit the reward function and guide the actor network to correctly output the optimal decision, thereby realizing dynamic operation and maintenance management and mode optimization of the production equipment system. Based on the above ideas, this step designs the following... Figure 8The framework for multi-objective operation mode optimization and operation and maintenance management of equipment systems is shown, and the design and implementation of the DRL optimization algorithm are further completed. The pseudocode of the DRL algorithm for adaptive optimization of equipment operation mode and operation and maintenance management scheme is as follows: Figure 10 and Figure 11 As shown.

[0256] This embodiment provides a multi-modal knowledge-driven multi-objective operation mode optimization method for production equipment systems. It introduces multi-modal knowledge graph technology to achieve the fusion of multi-modal data and knowledge. Based on this, it extends equipment performance status evaluation indicators to the applicability level, comprehensively incorporating production demand indicators (including output, time, quality, and energy consumption) into equipment performance evaluation to assess its applicability. Furthermore, it considers the combined effects of changes in production demand and system state, integrating various performance indicators such as basic performance, production capacity, task reliability, operational safety, output quality, maintenance cost-benefit, and operational energy consumption. This establishes a multi-index operation quality evaluation method for production equipment based on extended random flow networks. Under varying production demands, it rationally utilizes multiple performance indicators of production equipment, comprehensively integrating optimization strategies such as equipment operation mode control, predictive maintenance, and material and energy flow scheduling. It proposes a multi-objective adaptive equipment operation and maintenance management and mode optimization method, and introduces deep reinforcement learning technology to achieve dynamic optimization solutions for operation and maintenance strategies.

[0257] As a specific application embodiment of the present invention, for example Figure 12 The engineering case of the open-pit mine transportation equipment system shown illustrates the steps of this embodiment. In this system, vehicle trains are the main transportation tools in open-pit mining, and their main task is to transport the appropriate quantity and grade of ore from the mine to the crushing station according to the production needs of the mine and crushing station.

[0258] Step 1: Identification and collection of operational data from the open-pit mine transportation system. For example... Figure 13As shown in Figure (1), based on the engineering case of the open-pit mining transportation system, and according to the optimization requirements of the system's operation mode in the engineering, the multimodal operation data generated during its operation are identified, selected, and collected. Specifically, three vehicles in three vehicle groups are selected as application examples, numbered as equipment 1, 2, 3 (group 1), 4, 5, 6 (group 2), and 7, 8, 9 (group 3), respectively. The operation mode is optimized for them according to their different production needs and operating status in four production cycles. For the aforementioned equipment, the operational status data sources identified and selected in this step include four types of measurement data: equipment operating load data (engine speed, steering angle command and front wheel steering angle feedback, braking control quantity and braking feedback, throttle control quantity and throttle feedback), performance status observation data (unmanned driving lateral deviation, heading deviation, speed deviation), energy consumption status data (battery SOC), and material flow status data (operating speed, travel curvature and slope, material load), as well as three types of corpus data: equipment system fault diagnosis and maintenance records (including component fault codes), production demand and equipment operation task profile records, and equipment operation fault FMEA analysis report database. These data are then applied to the construction of the equipment multimodal knowledge graph (MKG).

[0259] Step 2: Data preprocessing for knowledge reasoning based on the triple association model. For example... Figure 13 As shown in (2), firstly, entity lists, relation lists, and triple lists are formed using entity information and relation information from a corpus database. Then, using... Figure 13 The “head entity-relationship-tail entity” triple list shown in (2) is used as input along with the aforementioned operating load data, equipment performance observation data, energy consumption status data, and material flow status data. Figure 13 The state data encoder in (3) performs vectorized encoding on each entity and relation. Specifically, in the entity and relation encoding here, except for the encoding vector of the equipment entity which comes from the operating data, the encoding of the other entities and relations all come from the text corpus information in the historical case library. Thus, a preliminary association model between the five types of entities, namely production equipment, operating parameters, production demand, optimization objectives and decision-making methods, can be constructed to complete the preprocessing of multimodal data.

[0260] Step 3: Constructing the MKG model of the device system based on multimodal operational data. For example... Figure 13 As shown in (3), the triplet list formed in the previous step and the aforementioned four types of quantitative data are used as input to perform vectorized encoding for each entity and relation. Based on this, as Figure 13As shown in (4), the multimodal information in the MKG weighted graph model is used as input, and a graph neural network is introduced to complete the overall encoding of MKG. Specifically, combining the multimodal knowledge and data in the operation of the open-pit mining and transportation system, the GNN used for the overall encoding of MKG, after training and convergence, contains the correlation weights between production demand, optimization objectives, and decision-making methods, such as... Figure 14 The heat map shown will be used in subsequent steps. Figure 15 The encoded result shown is incorporated into the state tensor of the Markov decision process. S t In this process, comprehensive integration of multimodal operational data and intelligent decision optimization are achieved.

[0261] Step 4: Apply the multimodal knowledge graph constructed in the previous step, based on... Figure 4 The process completes the identification and selection of key optimization objectives and key decision-making methods. Specifically, this step selects optimization objectives that meet the current needs from step S203, and correspondingly selects operation and maintenance decision-making and mode optimization methods that are suitable for the current needs from step S207, and constructs a joint optimization decision-making model.

[0262] Step 5: Based on the output of the previous step, the equipment operation and maintenance optimization goals can be identified and selected according to their importance. One or more of the 10 optimization goals proposed in step S203 are selected as the actual optimization goals for operation and maintenance, and then applied to the DRL algorithm. Specifically, based on the different production demand information within the four production cycles and applying MKG encoding weights, the selected optimization goals are shown in Table 1. Note that since this application example does not involve the simultaneous identification and selection of multiple optimization goals, this corresponds to the special case in step S207.

[0263] Step 6: Equipment Mode Optimization and Operation and Maintenance Management Decision Technology Selection. Based on the key objective identification results from Steps 6 and 7, one or more technologies with a high correlation to the key optimization objectives can be selected from the four proposed operation and maintenance decision technologies: equipment operation mode control strategy optimization, predictive maintenance strategy optimization, material flow scheduling strategy optimization, and energy flow scheduling strategy optimization. For the four production cycles of the equipment system in this application case, the selection results are shown in Table 1. Table 1 shows the optimization objectives and decision method selection results for the equipment system in different production cycles. By applying Markov Decision Process (MDP) to construct a joint strategy optimization model and configuring the DRL training environment, the DRL algorithm can be applied for joint optimization solutions.

[0264] Table 1

[0265]

[0266] Step 7: Applying the algorithm to problem solving. Based on the above application steps, the joint optimization algorithm for equipment mode optimization and operation and maintenance management strategies based on DRL proposed in this invention can be applied to jointly optimize and solve the operation and maintenance management and mode optimization strategies of the above-mentioned open-pit mining transportation equipment system. The training optimization convergence graph of the DRL algorithm for the four production cycles of the equipment system in this application case is attached. Figure 15 As shown in the figure, the proposed algorithm exhibits stable optimization performance. This preliminarily verifies the practicality of this method in an engineering example of open-pit mining equipment systems. For the four production cycles listed in Table 1, the results of equipment predictive maintenance optimization, energy flow scheduling, operation mode control, and material flow scheduling obtained by the DRL algorithm are shown in Table 2. Table 2 presents the equipment operation mode optimization strategies for each production cycle based on DRL joint optimization.

[0267] Table 2

[0268]

[0269] This embodiment also provides a multi-modal knowledge-driven multi-objective operation mode optimization device for a production equipment system. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0270] This embodiment provides a multi-modal knowledge-driven multi-objective operation mode optimization device for a production equipment system, including:

[0271] The first model building module is used to build a production demand change prediction model based on cyclical trend decomposition.

[0272] The second model building module is used to build a multimodal knowledge graph model of the production equipment system.

[0273] The third model construction module is used to construct an evaluation index system for the operation quality of production equipment systems, with the optimization objectives being basic equipment performance, remaining equipment lifespan, average maintenance interval, equipment operating time availability, equipment production capacity, equipment reliability, equipment operation safety, equipment output quality, equipment operation and maintenance costs and benefits, and equipment operation energy efficiency.

[0274] The fourth model building module is used to build a data-driven device operation model based on the extended random stream network model.

[0275] The evaluation module is used to evaluate various optimization objectives of the production equipment system based on the extended random flow network model.

[0276] The fifth model building module is used to construct a multi-objective reward function based on the knowledge-driven superior-inferiority distance method.

[0277] The sixth model building module is used to build a joint strategy optimization model for operation and maintenance management and mode optimization.

[0278] The identification and selection module is used to identify and select key optimization targets and operation and maintenance technologies driven by multimodal production needs and production equipment operation data based on multimodal knowledge graphs.

[0279] The optimization module is used to perform joint optimization of production equipment mode and operation and maintenance management strategy based on deep reinforcement learning and deep deterministic policy gradient.

[0280] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0281] In this embodiment, the multi-modal knowledge-driven production equipment system multi-objective operation mode optimization device is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0282] This invention also provides a computer device having the above-described multi-modal knowledge-driven multi-objective operation mode optimization device for production equipment systems.

[0283] Please see Figure 16 , Figure 16 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 16 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 16 Take a processor 10 as an example.

[0284] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0285] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0286] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0287] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0288] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 16 Taking the example of a connection between China and Israel via a bus.

[0289] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touch screen. Output device 40 may include a display device, etc.

[0290] This invention also provides a computer-readable storage medium in which the above-described methods can be implemented in hardware, firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and to be stored on a local storage medium, so that the methods described herein can be processed by such software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware.

[0291] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0292] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope of this application.

Claims

1. A multi-modal knowledge-driven method for optimizing the multi-objective operation mode of a production equipment system, characterized in that, The method includes: A production demand change prediction model is constructed based on cyclical trend decomposition. Construct a multimodal knowledge graph model for the production equipment system; Construct a quality assessment index system for production equipment system operation, with the optimization objectives being basic equipment performance, remaining equipment lifespan, average maintenance interval, equipment operating time availability, equipment production capacity, equipment reliability, equipment operating safety, equipment output quality, equipment operation and maintenance costs and benefits, and equipment operating energy efficiency. A data-driven device operation model is constructed based on the extended random stream network model. Based on the extended random flow network model, the various optimization objectives of the production equipment system are evaluated; Construct a multi-objective reward function based on a knowledge-driven method of distance between superior and inferior solutions; Construct a joint strategy optimization model for operation and maintenance management and mode optimization; Based on multimodal knowledge graphs, key optimization objectives and operation and maintenance technologies are identified and selected driven by multimodal production demand and production equipment operation data. Based on deep reinforcement learning, a deep deterministic policy gradient is used to jointly optimize production equipment mode and operation and maintenance management strategy. The construction of the joint strategy optimization model for operation and maintenance management and mode optimization includes: constructing an equipment operation mode control optimization model by adopting equipment operation mode control; constructing an equipment material flow scheduling optimization model by adopting equipment system material flow scheduling; constructing an energy flow scheduling optimization model by adopting equipment operation energy flow scheduling; and constructing an equipment predictive maintenance optimization model by adopting equipment predictive maintenance. The identification and selection of key optimization objectives and operation and maintenance technologies driven by multimodal production demand and production equipment operation data based on multimodal knowledge graphs includes: forming entity information, relationship lists, and triple lists based on entity information and relationship lists; encoding entities and relationships using triple lists and equipment operation data as input to obtain entity encoding vectors and relationship encoding vectors; calculating entity similarity and relationship similarity respectively using similarity calculation, assigning weights to the relationships between entities to form a weighted graph model; and ranking the optimization objectives and operation and maintenance technologies by importance based on the weighted graph model to select the optimization objectives and operation and maintenance technologies. The method, based on deep reinforcement learning and employing deep deterministic policy gradients, jointly optimizes production equipment operation modes and maintenance management strategies. This includes: constructing a Markov decision model for multi-objective maintenance management and mode optimization of production equipment, determining the state space, action space, and reward function of the decision-making process; designing an Actor-Critic network model as a deep reinforcement learning agent; and training the agent network using a deep deterministic measurement gradient algorithm to output production equipment operation mode optimization and maintenance management strategies.

2. The method according to claim 1, characterized in that, The construction of the multi-objective reward function based on the knowledge-driven superior-inferiority distance method includes: The optimization objectives for equipment operation modes are normalized. Extract nodes and association models from the multimodal knowledge graph, and extract the strength weight values ​​of the associations between various entities; Identify the activation nodes of the knowledge graph and calculate the weighted optimization objective vector; Calculate the Euclidean distance between the weighted optimization objective vector and the optimal and worst solutions in the objective space, and construct a multi-objective reward function.

3. The method according to claim 2, characterized in that, The process of determining the activation nodes of the knowledge graph and calculating the weighted optimization target vector includes: Based on the current equipment type, operating status, and production demand node category, determine the activation nodes of the knowledge graph; Starting from the activated node, traverse the predefined associated paths and calculate the importance of the optimization target according to the following formula; in, For the first i One optimization objective, For the first i The importance of each optimization objective From input to optimization goal Path weights; The importance of the optimization objective is applied to the objective function for weighted integration, resulting in a weighted optimization objective vector.

4. The method according to claim 2, characterized in that, The calculation of the Euclidean distance between the weighted optimization objective vector and the optimal and worst solutions in the objective space, and the construction of the multi-objective reward function, includes: The Euclidean distance between the optimal and worst solutions of the weighted optimization objective vector in the objective space is calculated using the following formula: in, To calculate the Euclidean distance of the optimal solution of the target vector in the target space using weighted optimization. To calculate the Euclidean distance of the worst solution of the objective vector in the objective space using weighted optimization. To optimize the target vector using weighted methods, The optimal solution is the zero vector. This is the worst solution; Construct a multi-objective reward function according to the following formula: in, This is a multi-objective reward function.

5. The method according to claim 1, characterized in that, With equipment operation safety as the optimization objective, the construction of the production equipment system operation quality evaluation index system includes: Assess the severity of the consequences of equipment failure and calculate the expected severity of the consequences of equipment failure; The failure frequency during equipment operation is assessed based on the proportional hazards regression model, and the failure probability during equipment operation is calculated. Assess the undetectability of equipment failure modes and calculate the expected value of the undetectability of equipment failure modes; Based on the expected severity of the consequences of equipment failure, the probability of failure during equipment operation, and the expected undetectability of equipment failure modes, a risk assessment model for the operation of production equipment systems is constructed.

6. The method according to claim 1, characterized in that, For time series The production demand change prediction model is as follows: in, This represents the trend term of long-term changes in the time series. Periodic components are used to characterize the local periodic features of a time series. This is the accidental effect term. This represents the random error in the model estimation results.

7. A computer device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the computer instructions to perform the multi-modal knowledge-driven multi-objective operation mode optimization method for production equipment systems as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the multi-modal knowledge-driven multi-objective operation mode optimization method for production equipment systems as described in any one of claims 1 to 6.

Images