A design method for existing building reinforcement guided by expert experience and rules

Abstract

This invention discloses a design method guided by expert experience and rules for the reinforcement of existing buildings, comprising: constructing a multi-level expert experience rule graph for the reinforcement of existing buildings; constructing an AI-generated model for the reinforcement design of existing buildings; coupling the expert experience rule graph to the AI-generated model via forward embedding; constructing a rule constraint mapping layer in the latent space of the AI-generated model; transforming the expert experience rules into differentiable rule constraint loss functions; jointly optimizing these loss functions with the original generation loss function and mechanical performance loss function of the AI-generated model to generate a reinforcement design scheme that conforms to expert experience constraints; refining and verifying the feasibility of the generated reinforcement design scheme at the node level; and outputting the final reinforcement design scheme. This invention solves the problems of existing reinforcement design AI-generated schemes lacking experience constraints and failing to meet engineering practice requirements, significantly improving the intelligence level and reliability of reinforcement design.

Description

Technical Field

[0001] This invention relates to the fields of building structure reinforcement technology and artificial intelligence technology, specifically to a design method and system guided by expert experience and rules for the reinforcement of existing buildings. Background Technology

[0002] With the increasing service life of urban buildings and the combined effects of natural disasters and environmental loads, the problems of structural damage, aging, and localized degradation are becoming increasingly prominent. Repair and reinforcement projects are playing an increasingly important role in the operation and maintenance of existing buildings. Many existing buildings in my country were constructed in the 1980s and 1990s, and generally suffer from problems such as chaotic structural systems, low material strength, and a lack of structural measures, resulting in their seismic performance failing to meet current code requirements. Therefore, scientific and efficient seismic reinforcement of these existing buildings is an urgent need in the current construction engineering field.

[0003] Traditional reinforcement design for existing buildings mainly relies on the following methods: Firstly, there is the manual experience-based design. Reinforcement schemes are typically drafted manually by senior structural engineers based on on-site inspection results, design specifications, and engineering experience, and are finalized after multiple scheme comparisons and structural calculations. This method heavily relies on the engineer's accumulated personal experience; different designers may provide significantly different reinforcement schemes, and the design cycle is long, inefficient, and makes it difficult to achieve overall optimization of the scheme.

[0004] Secondly, finite element numerical simulation-aided design. In recent years, structural design software such as PKPM and Yingjianke have successively developed assessment and reinforcement modules, which can perform load-bearing capacity verification and reinforcement scheme comparison for existing buildings. However, these software programs still rely on the human decision-making of engineers, and only provide calculation tools; they do not have the ability to autonomously generate and optimize reinforcement schemes.

[0005] Third, purely data-driven AI design methods. Some studies attempt to use deep learning neural networks for stress analysis and safety level assessment, combining graph neural networks and reinforcement learning to optimize reinforcement parameters. However, purely data-driven methods lack effective utilization of long-term accumulated empirical rules in engineering practice, and the design results often fail to meet the requirements of code construction and empirical constraints.

[0006] Fourth, digital twin and data fusion methods. Some patents disclose seismic reinforcement methods applicable to historical buildings, which integrate digital twin modeling and multi-objective optimization, but they do not systematically integrate expert experience rules in the field of engineering design into various stages of the intelligent design process.

[0007] In summary, how to systematically introduce the expert experience rules accumulated over a long period of engineering practice into the intelligent design process for strengthening existing buildings, so as to constrain and guide AI models to generate strengthening schemes that conform to engineering specifications, structural requirements and practical experience, is a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a design method guided by expert experience and rules for the reinforcement of existing buildings. By constructing a multi-level expert experience rule map for the reinforcement of existing buildings, and by embedding the experience rules in a forward manner to constrain and guide the generation process of AI reinforcement design models, this invention solves the technical problem that the AI-generated solutions in existing reinforcement designs lack experience constraints and do not meet the requirements of engineering practice.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A design methodology guided by expert experience and rules for the reinforcement of existing buildings includes the following steps: Step S1: Construct a multi-level expert experience rule map for the reinforcement of existing buildings. This specifically includes: S11 Collect and organize expert experience rules in the field of existing building reinforcement. The expert experience rules are derived from at least one of the following sources: mandatory and recommended clauses in structural design codes and reinforcement technical specifications, implicit design experience accumulated by senior structural engineers in long-term engineering practice, and historical case data of completed reinforcement projects.

[0010] S12 classifies and hierarchically organizes the collected expert experience rules, forming a rule map that includes at least three levels: Component-level regular layer: This refers to the geometric dimensional constraints, material strength constraints, and reinforcement constraints for individual components. For example, after adding a shear wall, the wall thickness should not be less than 140mm, and the minimum reinforcement ratio of vertical and horizontal distributed steel bars should not be less than 0.15%; the wall thickness of the bottom reinforced part of shear walls of seismic grades I and II should not be less than 200mm and should not be less than 1 / 16 of the story height.

[0011] Structural system-level rule layer: Constraint rules for the overall structural stress system, including constraints on the comprehensive seismic resistance index of the reinforced floors, and rules for determining the values ​​of the system influence coefficient and local influence coefficient. For example, the comprehensive seismic resistance index of the reinforced floors should not be less than 1.0.

[0012] Construction feasibility rules layer: Constraint rules for the feasibility and economy of reinforcement construction, including the priority of reinforcement method selection, feasibility of node construction, and construction space limitations.

[0013] S13 transforms the expert experience rules, after being categorized and hierarchically organized as described above, into a structured representation, forming a rule graph that can be read and invoked by a computer. The structured representation uses JSON format, and each rule contains five fields: rule identifier, applicable scenario conditions, rule preconditions, rule conclusion, and weight coefficient.

[0014] S14 Establish a dynamic rule update mechanism: When a new reinforcement design is completed and passes acceptance, the experience information of the new scheme is fed back into the rule graph. The weight coefficients and applicable conditions of the rule nodes are updated through incremental learning to achieve continuous iterative optimization of the expert experience rule graph.

[0015] Step S2: Construct an AI-generated model for the reinforcement design of existing buildings. This specifically includes: S21 collects existing building reinforcement design datasets. Each data entry includes the following information: original structural parameters of the existing building, inspection and assessment information, seismic fortification parameters, reinforcement design scheme, and structural performance indicators after reinforcement.

[0016] S22 Constructs an AI-based reinforcement design model based on a diffusion-generative network or a generative adversarial network. The input of the model is the original structural parameters and inspection and assessment information of the existing building, and the output is a preliminary reinforcement design scheme. The preliminary reinforcement design scheme includes at least: reinforcement location identification, reinforcement method type, geometric dimensions of the reinforcement components, and material parameters.

[0017] Step S3: Couple the expert experience rule graph constructed in Step S1 to the AI ​​generative model constructed in Step S2 via forward embedding to generate a reinforcement design scheme that conforms to expert experience constraints. Specifically, this includes: S31 constructs a rule constraint mapping layer in the latent space of the AI ​​generative model, transforming expert experience rules into differentiable rule constraint loss functions. The form of the rule constraint loss function for each rule is as follows: L rule k =max(0,violation k ) Among them, violence k Indicates the first k The degree of violation of a rule of expert experience is quantified; when the generated solution satisfies the rule, the violation occurs. k =0, the greater the degree of violation, the higher the violation rate. k The larger.

[0018] S32 defines the total loss function of rule constraints as the weighted sum of the loss functions of each rule constraint: L total_rule =k∑wk⋅Lrule k in, wk For the first k The weight coefficient of each rule is determined based on the importance and mandatory level of the rule.

[0019] S33 jointly optimizes the total loss function of rule constraints with the original generation loss function and mechanical performance loss function of the AI ​​generation model. The joint loss function takes the form of: Ljoint=λdata⋅Ldata+λmech⋅Lmech+λrule⋅Ltotal_rule in, Ldata For data fitting loss, Lmech For mechanical property loss, Ltotal_rule The total loss is constrained by the rules. λdata , λmech and λrule These are the weighting coefficients for the corresponding losses.

[0020] In each iteration of the AI-generated model, S34 simultaneously calculates data loss, mechanical loss, and rule constraint loss via forward propagation, and updates model parameters through backpropagation. This allows the model to gradually learn generation strategies that satisfy expert experience rules during training, ensuring that the final reinforcement design scheme simultaneously satisfies data fit, mechanical rationality, and compliance with experience rules.

[0021] Step S4: Refine the reinforcement design scheme generated in Step S3 at the node level and verify its feasibility. Specifically, this includes: S41 refines the node construction design of the generated reinforcement design scheme, including the connection node construction between the reinforcement component and the existing component, the lap joint and anchorage construction between the reinforcement components.

[0022] S42 calls the construction feasibility verification module to verify the construction feasibility of the plan according to the rules in the construction feasibility rule layer. For plans that do not meet the construction feasibility rules, it automatically makes local adjustments or triggers regeneration.

[0023] S43 outputs the final reinforcement design scheme that satisfies all expert experience rules constraints. The final reinforcement design scheme is presented in the form of structural diagrams, parameter tables, and construction instructions.

[0024] Furthermore, in step S1, the structured representation of the expert experience rules adopts the form of a multi-level rule graph, including the following components: Rule node: Represents a specific expert experience rule, which includes five attributes: rule ID, rule name, rule type, rule content, and rule source; Scene node: Represents the building scene to which the rule applies, including attributes such as building type, seismic fortification intensity, building height, structural system, and subsequent service life; Related edges: represent the applicability relationship between rule nodes and scene nodes, as well as the conflict, dependency, and priority relationships between rule nodes; Conflict resolution module: Used to detect possible rule conflicts in the rule graph and automatically eliminate conflicts based on preset conflict resolution strategies (based on the highest priority rule, the stricter rule, or the newest version of the rule).

[0025] Furthermore, in step S3, the forward embedding method can also adopt a rule-driven generation space constraint method: the expert experience rules are transformed into geometric constraint boundaries and parameter value range constraints in the generation space. In each iteration of the AI ​​generation model generation scheme, the generation parameters are limited within the rule constraint boundaries, so that the generation scheme naturally satisfies the experience rules, rather than performing rule violation corrections after generation.

[0026] Furthermore, the method also includes a multi-objective optimization module: under the premise of satisfying the constraints of expert experience rules, the reinforcement design scheme generated in step S3 is optimized with at least one of the following objectives: lowest reinforcement cost, shortest construction period, and least impact on building use, and the Pareto optimal solution set is output for designers to choose from. Attached Figure Description

[0027] Figure 1 The flowchart of the method of the present invention shows the main process of steps S1 to S4.

[0028] Figure 2 This invention describes the process of constructing a multi-level expert experience rule graph.

[0029] Figure 3 This describes the data acquisition and AI model construction process of this invention.

[0030] Figure 4 This is the mathematical form of the forward embedding coupling mechanism and the joint loss function of this invention.

[0031] Figure 5 This is the process of refining and verifying the nodes in this invention. Detailed Implementation

[0032] The present invention will now be described in detail with reference to specific embodiments. These embodiments are for illustrative purposes only and do not constitute a limitation on the scope of protection of the present invention.

[0033] Example 1: Intelligent design for seismic reinforcement of a five-story frame office building This embodiment takes a five-story reinforced concrete frame structure office building built in 1995 as an example. The building is located in an area with a seismic fortification intensity of 8 degrees (0.20g), a seismic fortification category of Class C, and a subsequent service life of 40 years. On-site inspection and assessment revealed that the building has problems such as excessive axial compression ratios in some frame columns, insufficient reinforcement in frame beams, and missing connection structures between infill walls and frames. The intelligent design method of this invention is now used to design seismic reinforcement for this building.

[0034] Step S1: Based on the "Standard for Seismic Appraisal of Buildings" GB 50023, the "Technical Specification for Seismic Strengthening of Buildings" JGJ116, and the engineering experience of senior structural engineers, collect and organize expert experience rules in the field of strengthening, forming a three-level rule map including component level, structural system level, and construction feasibility level. Some rule examples are as follows: Rule R001 (Component Level - Beam): When reinforcing a frame beam by increasing the cross section, the thickness of the new cross section shall not be less than 60mm, and the strength grade of the new concrete shall not be lower than the strength grade of the original beam concrete and shall not be lower than C25.

[0035] Rule R002 (Component Level - Column): When reinforcing frame columns using the external steel cladding method, the side length of the angle steel section should not be less than 75mm, and the spacing of the steel gusset plates should not be greater than 40 times the radius of gyration of the single-limb angle steel section and should not be greater than 400mm.

[0036] Rule R003 (structural system level): The comprehensive seismic resistance index of the floors of the reinforced structure shall not be less than 1.0 and shall not exceed 20% of the index value of the adjacent floors.

[0037] Step S2: Collect data from 1000 existing building reinforcement design cases and construct an AI reinforcement design model based on a diffusion generative network. The model's input feature vector includes 48 parameters: building height, number of stories, structural system, seismic fortification intensity, site category, concrete strength test value, rebar configuration test value, axial compression ratio exceeding limit rate, and insufficient reinforcement rate. The model's output is a reinforcement design scheme vector.

[0038] Step S3: Transform the rules in the rule graph into differentiable rule constraint loss functions. Taking rule R001 as an example, the corresponding loss function is: L R001=max(0,60− d new)+max(0, C original− C new) in d new represents the newly added cross-sectional thickness (mm). C "new" refers to a newly added concrete strength grade. C`original` refers to the original concrete strength grade of the beam. The weighting coefficients of the joint loss function are set to... λ data=0.3、 λ mech=0.4 λ rule=0.3. After 200 rounds of training, the proportion of model-generated solutions that satisfy all rule constraints increased from the initial 32% to 96%.

[0039] Step S4: The optimal reinforcement scheme generated by the model is as follows: The frame columns from the ground floor to the third floor are reinforced using the increased cross-section method, with an additional cross-section thickness of 100mm and a concrete strength grade of C30; the frame columns from the fourth to fifth floors are reinforced using the external steel cladding method, employing L80×8 angle steel and 8mm thick steel gusset plates; all frame beams are reinforced using the carbon fiber reinforcement method, employing double-layer bonding of 300g / ㎡ carbon fiber cloth. The scheme has been verified for construction feasibility and meets the requirements for construction space and joint construction. PKPM software calculations show that the minimum comprehensive seismic resistance index of each floor after reinforcement is 1.12, meeting the code requirements. Compared to traditional manual design, this method shortens the design cycle from approximately 10 working days to 0.5 working days, and reduces the total reinforcement cost by approximately 21%.

[0040] Example 2: Verification of the dynamic rule update mechanism After completing Example 1, the actual reinforcement design scheme, experience feedback during construction, and post-reinforcement testing data were used as new empirical data to update the rule graph through incremental learning. Specific updates included: based on the experience that L80×8 angle steel is more commonly used than L75×8 angle steel in actual construction, the weight coefficient of "the side length of the angle steel section should not be less than 75mm" in rule R002 was increased from 0.8 to 0.9; based on post-reinforcement testing data, the upper limit of the steel gusset spacing in the external steel method was adjusted from 40 times the section radius of gyration to 45 times the section radius of gyration, and the rule expression was updated accordingly. The updated rule graph was applied to the design of subsequent similar projects, further improving design quality and customer satisfaction.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A design method oriented to reinforcement of existing buildings guided by expert experience and rules, characterized in that, Includes the following steps: Step S1: Construct a multi-level expert experience rule map for the reinforcement of existing buildings. The expert experience rule map includes at least three levels: component-level rule layer, structural system-level rule layer, and construction feasibility rule layer, and is stored in a structured form. Step S2: Construct an AI-generated model for the reinforcement design of existing buildings. The AI-generated model takes the original structural parameters and inspection and assessment information of the existing buildings as input and outputs a preliminary reinforcement design scheme. Step S3: Couple the expert experience rule graph constructed in Step S1 to the AI ​​generation model constructed in Step S2 using a forward embedding method to generate a reinforcement design scheme that conforms to expert experience constraints. The forward embedding method refers to constructing a rule constraint mapping layer in the latent space of the AI ​​generation model, transforming the expert experience rules into a differentiable rule constraint loss function, and jointly optimizing it with the original generation loss function and mechanical performance loss function of the AI ​​generation model, so that the AI ​​generation model is simultaneously guided and constrained by expert experience rules during training and generation. Step S4: Refine the reinforcement design scheme generated in step S3 at the node level and verify its feasibility, and output the final reinforcement design scheme.

2. The method of claim 1, wherein, In step S3, the joint optimization of the rule constraint loss function is achieved through the following joint loss function: L joint= λ data L data+ λ mech L mech+ λ rule k ∑ wkL rule, k in, L data represents the data fitting loss. L mech represents the loss of mechanical properties. L rule k Let k be the rule constraint loss function corresponding to the k-th expert experience rule. wk For the first k The weight coefficient of each rule, λ data λ mech and λ The rule is the weighting coefficient for the corresponding loss.

3. The method according to claim 2, characterized in that, The rule constraint loss function corresponding to the kth expert experience rule is: L rule, k =max(0,violation k ) violation k is a quantified value representing the degree of violation of the rule, violation k = 0 when the generated plan satisfies the rule.

4. The method of claim 1, wherein, The multi-level expert experience rule graph in step S1 includes rule nodes, scene nodes, and associated edges; the rule node includes five attributes: rule ID, rule name, rule type, rule content, and rule source; the scene node includes attributes such as building type, seismic fortification intensity, building height, structural system, and subsequent service life; the associated edges are used to represent the applicability relationship between rule nodes and scene nodes, as well as the conflict relationship, dependency relationship, and priority relationship between rule nodes.

5. The method of claim 1, wherein, Step S1 also includes a dynamic rule update mechanism: when a new reinforcement design is completed and passes acceptance, the experience information of the new scheme is fed back into the rule graph, and the weight coefficients and applicable conditions of the rule nodes are updated through incremental learning to achieve continuous iterative optimization of the expert experience rule graph.

6. The method of claim 1, wherein, The forward embedding method in step S3 also includes a rule-driven generative space constraint method: the expert experience rules are transformed into geometric constraint boundaries and parameter value range constraints in the generative space, and the generative parameters are limited to the rule constraint boundaries in each iteration of the AI ​​generative model generative scheme.

7. The method of claim 1, wherein, The multi-level expert experience rule graph also includes a conflict resolution module, which is used to detect possible rule conflicts in the rule graph and automatically eliminate conflicts based on a preset conflict resolution strategy; the conflict resolution strategy includes at least one of the following: using the highest priority rule, using a more stringent rule, and using the latest version of the rule.

8. The method according to claim 1, characterized in that, The method also includes a multi-objective optimization module: under the premise of satisfying the constraints of expert experience rules, the reinforcement design scheme generated in step S3 is optimized with the goal of minimizing reinforcement cost, shortest construction period, and least impact on building use, and the Pareto optimal solution set is output.

9. The method of claim 1, wherein, Step S4 includes: refining the node structure design of the generated reinforcement design scheme, and calling the construction feasibility verification module to verify the construction feasibility of the scheme.

10. A rule-based expert system for intelligent design of reinforcement of existing buildings, characterized in that, include: The rule graph construction module is used to perform step S1 as described in claim 1; The AI-generated model module is used to perform step S2 as described in claim 1; A rule forward embedding module is used to perform step S3 as described in claim 1; The scheme output and verification module is used to execute step S4 as described in claim 1.

Images