World model based labmate scenario generation and multi-dimensional failure attribution method

Abstract

The application discloses a world model-based unmanned laboratory scheme generation and multi-dimensional failure attribution method, and belongs to the technical field of laboratory automation and intelligent system; comprising S1, constructing an experimental world model; S2, generating a candidate experimental scheme based on the experimental world model, and performing constraint filtering and comprehensive sorting on the candidate scheme to select an optimal scheme; S3, performing execution flow arrangement on the selected optimal scheme; S4, performing an experimental flow according to the flow arrangement result, and simultaneously performing real-time monitoring; S5, performing multi-dimensional verification on an experimental result, and generating a verification report; S6, checking whether there is an experimental failure and a result not meeting an expectation according to the verification report, and if yes, performing multi-dimensional failure attribution, otherwise, performing S7; S7, performing model backwriting and cyclic evolution; by adopting the method, the application can uniformly express experimental full elements, systematically generate an experimental scheme, multi-dimensionally verify an experimental result, and perform structured failure attribution.

Description

Technical Field

[0001] This invention belongs to the field of laboratory automation and intelligent systems technology, and in particular relates to a method for generating unmanned laboratory solutions and multidimensional failure attribution based on a world model. Background Technology

[0002] Unmanned laboratories utilize robots, automated instruments, and intelligent algorithms to complete experimental processes, improving experimental efficiency and repeatability. Existing unmanned laboratories often focus on automating experimental actions, instrument scheduling, or optimizing single-point parameters, typically lacking a unified expression of experimental objectives, instrument status, environmental conditions, researcher intentions, and failure feedback.

[0003] In materials research and development, biological experiments, chemical experiments, and engineering testing, experimental systems not only need to execute experiments, but also understand experimental objectives, propose candidate solutions, assess risks, verify the reliability of results, and provide explainable suggestions for the next round of adjustments after failures. If the system cannot incorporate the researcher's intentions, the state of the experimental subjects, and the state of the instruments into the same world model, it will be difficult to upgrade from "automatic execution" to "autonomous design."

[0004] The specific shortcomings of the existing solutions include: First, experimental design generation is mostly based on preset templates or univariate searches, lacking the ability to generate systematic designs based on experimental objectives and failing to flexibly switch between hypothesis-driven, analogical reasoning, and exploratory searches. When experimental objectives are complex or the parameter space has high dimensionality, preset templates cannot cover the effective search space.

[0005] Second, the execution process lacks a joint consideration of instrument status, reagent constraints, and environmental conditions, leading to frequent resource conflicts or constraint violations during the execution phase. Experimental protocols may appear feasible during the generation phase, but fail during execution due to unavailable instruments, incompatible reagents, or unmet environmental conditions.

[0006] Third, the verification of experimental results lacks a multi-dimensional assessment of the reliability of the results. It only compares the magnitude of the results with the expected values, without considering instrument accuracy, environmental interference, and statistical significance. When the instrument accuracy is insufficient or there is environmental interference, even if the experiment is performed correctly, the results may still be unreliable.

[0007] Fourth, the lack of systematic attribution after experimental failures makes it impossible to distinguish between hypothesis errors, execution biases, instrument malfunctions, and environmental interference, resulting in unclear directions for improvement in the next round of experiments. Existing methods typically attribute all failures simply to "experimental failure" without analyzing the root causes.

[0008] Fifth, the experimental experience lacks structured accumulation, making it unsuitable for supporting the optimization of subsequent experimental plans and the transfer of knowledge across experiments. The experience gained from each experiment is stored in the form of unstructured logs, which cannot be effectively utilized in subsequent experiments. Summary of the Invention

[0009] The purpose of this invention is to provide a method for generating unmanned laboratory schemes and attributing failures in a multidimensional manner based on a world model. This method can uniformly express all elements of an experiment, systematically generate experimental schemes, verify experimental results in multiple dimensions, and perform structured failure attribution.

[0010] To achieve the above objectives, this invention provides a method for generating unmanned laboratory solutions and attributing failures in a multidimensional manner based on a world model, comprising the following steps: S1. Construct an experimental world model; S2. Based on the experimental world model, generate candidate experimental schemes, perform constraint filtering and comprehensive ranking on the candidate schemes, and select the optimal scheme. S3. Arrange the execution flow for the selected optimal solution; S4. Conduct the experimental procedure according to the programmed results, and monitor it in real time. S5. Perform multidimensional verification of the experimental results and generate a verification report; S6. Based on the verification report, check whether there are any experimental failures or results that do not meet expectations. If so, perform multidimensional failure attribution; otherwise, proceed to S7. S7. Perform model write-back and iterative evolution.

[0011] Preferably, in S1, the experimental world model is obtained by constructing a target layer, an object layer, an instrument layer, an environment layer, a knowledge layer, and a researcher intent layer. ; S11. Construct the target layer Including research objectives Scientific Hypotheses Expected Results Successful judgment criteria and priority ; S12, Constructing the object layer Including the state of experimental materials Sample properties Reaction kinetic parameters and historical experimental data ; S13, Constructing the Instrument Layer Including instrument availability Accuracy status Calibration status Operational constraints and maintenance plan ; Define the instrument accuracy model The process is as follows: ; in For instruments The systematic bias, For instruments The standard deviation of random error This represents the true value of the measurement. Indicator Instrument The random error term; S14, Constructing the Environment Layer Including temperature ,humidity Cleanliness ,vibration and electromagnetic environment ; Define the environment state vector The process is as follows: ; S15, Building the Knowledge Layer Including domain knowledge graphs Known patterns Similar experimental experience and safety constraints ; S16. Constructing the Researcher Intent Layer Including researchers' goal statements Constraint preferences Risk tolerance and historical revision records ; S17. Establish an experimental world model based on the six levels constructed in S11-S16. The process is as follows: .

[0012] Preferably, the process of S2 is as follows: S21. Conduct hypothesis-driven generation based on current scientific hypotheses. Derive a verifiable experimental design; identify the independent variables. Dependent variable and control variables At the same time, the solution must meet the requirement of identifiability; S22. Perform analogical reasoning to generate successful experiences of similar experiments in the knowledge layer, and define experimental similarity. The process is as follows: ; in For similar experiments and Target similarity, For similar experiments and The similarity of the experimental subjects For similar experiments and Similarity of experimental conditions These are the corresponding similarity weights; Select the one with the highest similarity We extracted successful solutions from historical experiments as analogies and adapted them to the current experimental conditions. S23. Perform exploratory generation: Generate exploratory solutions in the parameter space based on Bayesian optimization, with the goal of maximizing information gain. The process is as follows: ; in Indicates the data acquisition function. This indicates the parameter values ​​for the next exploratory experiment. Represents historical data; S24. Generate a set of candidate experimental schemes based on the three strategies in S21-S23. ; S25. On the set of candidate solutions Perform constraint filtering to filter out infeasible solutions. The process is as follows: ; in Indicates a feasible solution. For instrument availability checks, For reagent compatibility testing, For safety constraint checks, For environmental condition inspection; S26. Rank the feasible solutions comprehensively and define the ranking criteria. : ; in For the plan The expected information gain, For execution costs, For risk level, To ensure a good match with the researchers' intentions, , , and These are the weighting coefficients for expected information gain, execution cost, risk level, and the degree of matching of researchers' intentions, respectively. S27. Based on the ranking results, find the solution with the highest score, i.e., the optimal solution. ; .

[0013] Preferably, the acquisition function in S23 is either the expected improvement function or the knowledge gradient function.

[0014] Preferably, the process of S3 is as follows: S31. Decompose the task into a sequence of atomic operations. Each atomic operation Defined as a quintuple; ; in For operation type, For operation parameters, For the expected duration, For the required resources, As a prerequisite; S32. Perform dependency analysis and construct an operation dependency graph. , among which the side Indicates operation Dependent on operation The completion; Dependency types include time-series dependencies, resource dependencies, and data dependencies; S33. Perform parallel optimization: based on the dependency graph, perform topological sorting to identify the set of operations that can be executed in parallel, and calculate the shortest execution time using the critical path method. : ; in Indicates the expected duration; S34. Conduct risk assessment: assess the risk of each operational step and calculate the probability of failure. The process is as follows: ; in This represents the historical failure rate for this type of operation. For operational complexity, The current health status of the instruments used. This represents the failure probability bias term. , and These represent the weighting coefficients corresponding to historical failure rate, operational complexity, and the current health status of the instrument used; S35. Develop contingency plans for situations where the probability of failure exceeds a preset failure threshold. The key steps include developing contingency plans and rollback strategies for failures. S36. Perform resource scheduling, coordinate instrument usage time, reagent consumption, and environmental control requirements, and avoid resource conflicts through constraint satisfaction problem solving.

[0015] Preferably, the process of S4 is as follows: S41. During execution, process monitoring is performed, and each operation step is monitored. Monitor the deviation between its actual execution status and the expected status. ; ; in Indicates the actual execution status. Indicates the expected state; when An abnormal alarm is triggered at any time. This indicates the preset abnormal threshold. S42. Anomaly detection based on statistical process control methods The process is as follows: ; in and These are the mean and standard deviation of the process variable, respectively. S43. Make a decision on the handling of the situation, and decide on the corresponding handling of the anomaly, such as continuing execution, pausing and waiting, adjusting parameters, or terminating the experiment, based on the type and severity of the anomaly. S44. Completely record all operations, observations, and decisions during the experiment to form a traceable experimental log.

[0016] Preferably, the process of performing multidimensional verification of the experimental results and generating a verification report in S5 is as follows: S51. Conduct target conformity verification and calculate the deviation between the experimental results and the expected target. The process is as follows: ; in Indicates the experimental results, Indicates the expected goal; when The target is determined to be compatible at that time. This indicates the preset deviation threshold. S52. Perform hypothesis testing. Determine whether the results support or refute the original hypothesis through hypothesis testing, and calculate the test statistic. : ; in This represents the mean of the experimental results. The expected value under the assumption, The standard deviation of the sample is 1. This represents the number of times the experiment was repeated. when At the significance level We reject the null hypothesis. This indicates a preset statistical threshold. S53. Conduct a credibility assessment and comprehensively evaluate the credibility of the results. ; ; in To ensure the accuracy and reliability of the instrument, For environmental stability and credibility, To assess statistical significance and reliability, For repeatability reliability, , , and These represent the corresponding credibility weight coefficients; S54. Perform a consistency check, comparing the results with known patterns. and historical data Perform a consistency check ; ; in It serves as a distance metric between the results and known patterns. S55. Generate a structured verification report containing verification conclusions, confidence levels, and evidence based on the multidimensional verification results.

[0017] Preferably, the process of multidimensional failure attribution in S6 is as follows: S61. Perform hypothesis attribution, determine whether the scientific hypothesis itself is flawed, and calculate the degree of contradiction between the hypothesis and the results. The process is as follows: ; When the degree of contradiction exceeds the preset contradiction threshold When the hypothesis is established, determine that the attribution is valid and analyze the direction in which the hypothesis needs to be revised. S62. Perform execution attribution to determine if there were any operational deviations during the execution process; and backtrack the execution records to calculate the execution deviation index. ; ; in and The first Actual and planned parameters for each step of the operation; when When the attribution for execution is established, This indicates the preset execution deviation threshold; S63. Perform instrument attribution to determine if there is an instrument malfunction or insufficient accuracy; check instrument status records and calibration data, and calculate the contribution of instrument deviation. : ; in This is due to instrument system deviation. For random error, Confidence coefficient; when When the instrument attribution is established, This indicates the preset instrument deviation threshold. S64. Perform environmental attribution, determine whether environmental conditions exceed expectations, and calculate the environmental deviation index. : ; in Indicates actual environmental conditions. Indicates expected environmental conditions. Represents a set of time points; when When the environmental attribution is deemed valid, This indicates the preset environmental deviation threshold; S65. Perform reagent attribution to determine if there are any reagent quality issues; check reagent batches, storage conditions, and expiration dates, and calculate reagent quality fractions. : ; in The passing score for the batch. For storage condition compliance score, For the validity period compliance score, , and These represent the corresponding compliance score weighting coefficients; when When the attribution of the reagent is deemed valid, This indicates the preset reagent quality threshold; S66. Perform attribution synthesis judgment, calculate the attribution confidence of each dimension, and generate attribution vectors. ; ; in These represent the confidence levels of the corresponding dimension attributions in S61 to S66, respectively. S67. Generate a structured attribution report, including attribution conclusions, chain of evidence, and improvement recommendations; S68. Based on the attribution results, generate adjustment suggestions for the next round of experiments.

[0018] Preferably, the process of S7 is as follows: S71. Update the property parameters and dynamic model of the experimental object; ; S72. Update the instrument's accuracy model and reliability model; S73. Accumulate experimental experience and update knowledge. \mathcal{L}_{\text{know}} ; S74, Parameters of the optimization scheme generation strategy; S75. Update the threshold of the risk prediction model; S76. Generate an experimental evolution log to record the inputs, outputs, conclusions, and improvements for each round of experiments; S77. Return to step S2 and proceed to the next experimental cycle.

[0019] Preferably, the unmanned laboratory scheme generation and multidimensional failure attribution method are implemented through a corresponding system; the system includes a candidate experimental scheme generation module for generating candidate schemes through three strategies: hypothesis-driven, analogical reasoning, and exploratory. Constraint filtering and ranking module used for feasibility filtering and comprehensive ranking of candidate solutions; An execution flow orchestration module used for task decomposition, dependency analysis, parallel optimization, and resource scheduling of selected solutions; An experimental execution and real-time monitoring module used to execute experimental procedures and monitor process status; A multi-dimensional verification module for performing four-dimensional verification of experimental results: target conformity, hypothesis testing, credibility, and consistency. A multidimensional failure attribution module for attributing experimental failures across five dimensions: hypothesis, execution, instrument, environment, and reagents. The Model Writeback and Loop Evolution module is used to write back experimental results and attribution analysis to the experimental world model.

[0020] Therefore, the above-mentioned method for generating unmanned laboratory schemes and attributing failures based on world models, adopted in this invention, has the following beneficial effects: (1) Through the six-layer experimental world model, a unified expression of all experimental elements is realized, enabling the experimental system to have the ability to "understand the experiment"; (2) By flexibly combining three scheme generation strategies (hypothesis-driven, analogical reasoning, and exploratory), a systematic scheme generation was achieved, covering both verification experiments and exploratory experiments. (3) By using constraint filtering and execution process orchestration, ensure the feasibility of the solution during the execution phase and avoid the problem of "design is feasible but execution fails"; (4) Verify the results in four dimensions (target compliance, hypothesis testing, credibility, and consistency) to ensure the reliability of the experimental conclusions; (5) By using five-dimensional failure attribution (hypothesis, execution, instrument, environment, reagent), the causes of failure can be systematically identified, providing clear directions for improvement in the next round of experiments; (6) Through model rewriting and iterative evolution, we can achieve continuous accumulation of experimental knowledge and continuous improvement of experimental capabilities.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a flowchart of the unmanned laboratory scheme generation and multidimensional failure attribution method based on the world model of the present invention. Figure 2 This is a system implementation diagram of the unmanned laboratory scheme generation and multidimensional failure attribution method based on the world model of the present invention. Detailed Implementation

[0023] 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Specific model specifications need to be selected and determined according to the actual specifications of the device, etc. The specific selection calculation method adopts existing technology in the art, and therefore will not be described in detail.

[0024] Example 1; like Figure 1 and Figure 2 As shown, this invention provides a method for generating unmanned laboratory solutions and attributing failures in multiple dimensions based on a world model, comprising the following steps: S1. Construct an experimental world model; The experimental world model is obtained by constructing the target layer, object layer, instrument layer, environment layer, knowledge layer, and researcher intent layer. ; S11. Construct the target layer Including research objectives Scientific Hypotheses Expected Results Successful judgment criteria and priority ; Scientific hypotheses are formally represented as follows: ; in Represents the vector of independent variables. Indicates the dependent variable. For parameters to be verified, This is experimental noise; S12, Constructing the object layer Including the state of experimental materials Sample properties Reaction kinetic parameters and historical experimental data ; Reaction kinetic parameters are characterized by kinetic models; ; in For concentration vectors, It is a function of the reaction rate. For temperature, For pressure; S13, Constructing the Instrument Layer Including instrument availability Accuracy status Calibration status Operational constraints and maintenance plan ; Define the instrument accuracy model The process is as follows: ; in For instruments The systematic bias, For instruments The standard deviation of random error This represents the true value of the measurement. Indicator Instrument The random error term; S14, Constructing the Environment Layer Including temperature ,humidity Cleanliness ,vibration and electromagnetic environment ; Define the environment state vector The process is as follows: ; S15, Building the Knowledge Layer Including domain knowledge graphs Known patterns Similar experimental experience and safety constraints ; S16. Constructing the Researcher Intent Layer Including researchers' goal statements Constraint preferences Risk tolerance and historical revision records ; S17. Establish an experimental world model based on the six levels constructed in S11-S16. The process is as follows: .

[0025] S2. Based on the experimental world model, generate candidate experimental schemes, perform constraint filtering and comprehensive ranking on the candidate schemes, and select the optimal scheme. S21. Conduct hypothesis-driven generation based on current scientific hypotheses. Derive a verifiable experimental design; identify the independent variables. Dependent variable and control variables At the same time, the solution must meet the requirement of identifiability; ; in, The dimension of the parameter to be verified is given by this condition, which ensures that the experimental data are sufficient to identify all the parameters to be verified. S22. Perform analogical reasoning to generate successful experiences of similar experiments in the knowledge layer, and define the experimental similarity. The process is as follows: ; in For similar experiments and Target similarity, For similar experiments and The similarity of the experimental subjects For similar experiments and Similarity of experimental conditions These are the corresponding similarity weights; Select the one with the highest similarity We extracted successful solutions from historical experiments as analogies and adapted them to the current experimental conditions. S23. Perform exploratory generation: Generate exploratory solutions in the parameter space based on Bayesian optimization, with the goal of maximizing information gain. The process is as follows: ; in Indicates the data acquisition function. This indicates the parameter values ​​for the next exploratory experiment. Represents historical data; The acquisition function can be either the expected improvement function or the knowledge gradient function.

[0026] Expected Improvement is used as the acquisition function: ; in, This is the optimal target value observed so far; Knowledge gradient is used as the data collection function: ; in Let be the mean function of the Gaussian process after the nth observation; S24. Generate a set of candidate experimental schemes based on the three strategies in S21-S23. ; S25. On the set of candidate solutions Perform constraint filtering to filter out infeasible solutions. The process is as follows: ; in Indicates a feasible solution. For instrument availability checks, For reagent compatibility testing, For safety constraint checks, For environmental condition inspection; S26. Rank the feasible solutions comprehensively and define the ranking criteria. : ; in For the plan The expected information gain, For execution costs, For risk level, To ensure a good match with the researchers' intentions, , , and These are the weighting coefficients for expected information gain, execution cost, risk level, and the degree of matching of researchers' intentions, respectively. S27. Based on the ranking results, find the solution with the highest score, i.e., the optimal solution. ; .

[0027] S3. Arrange the execution flow for the selected optimal solution; S31. Decompose the task into a sequence of atomic operations. Each atomic operation Defined as a quintuple; ; in For operation type, For operation parameters, For the expected duration, For the required resources, As a prerequisite; S32. Perform dependency analysis and construct an operation dependency graph. , among which the side Indicates operation Dependent on operation The completion; Dependency types include time-series dependencies ( Must Execute after completion), resource dependencies ( and Sharing the same resources) and data dependencies The input is (output) S33. Perform parallel optimization: based on the dependency graph, perform topological sorting to identify the set of operations that can be executed in parallel, and calculate the shortest execution time using the Critical Path Method (CPM). : ; in Indicates the expected duration; S34. Conduct risk assessment: assess the risk of each operational step and calculate the probability of failure. The process is as follows: ; in This represents the historical failure rate for this type of operation. For operational complexity, The current health status of the instruments used. This represents the failure probability bias term. , and These represent the weighting coefficients corresponding to historical failure rate, operational complexity, and the current health status of the instrument used; S35. Develop contingency plans for situations where the probability of failure exceeds a preset failure threshold. The key steps include developing contingency plans and rollback strategies for failures. S36. Perform resource scheduling, coordinate instrument usage time, reagent consumption, and environmental control requirements, and avoid resource conflicts by solving constraint satisfaction problems (CSP).

[0028] S4. Conduct the experimental procedure according to the programmed results, and monitor it in real time. S41. During execution, process monitoring is performed, and each operation step is monitored. Monitor the deviation between its actual execution status and the expected status. ; ; in Indicates the actual execution status. Indicates the expected state; when An abnormal alarm is triggered at any time. This indicates the preset abnormal threshold. S42. Anomaly detection based on statistical process control methods The process is as follows: ; in and These are the mean and standard deviation of the process variable, respectively. S43. Make a decision on the handling of the situation, and decide on the corresponding handling of the anomaly, such as continuing execution, pausing and waiting, adjusting parameters, or terminating the experiment, based on the type and severity of the anomaly. S44. Completely record all operations, observations, and decisions during the experiment to form a traceable experimental log.

[0029] S5. Perform multidimensional verification of the experimental results and generate a verification report; The process of performing multidimensional verification of the experimental results and generating a verification report is as follows: S51. Conduct target conformity verification and calculate the deviation between the experimental results and the expected target. The process is as follows: ; in Indicates the experimental results, Indicates the expected goal; when The target is determined to be compatible at that time. This indicates the preset deviation threshold. S52. Perform hypothesis testing. Determine whether the results support or refute the original hypothesis through hypothesis testing, and calculate the test statistic. : ; in This represents the mean of the experimental results. The expected value under the assumption, The standard deviation of the sample is 1. This represents the number of times the experiment was repeated. when At the significance level We reject the null hypothesis. This indicates a preset statistical threshold. S53. Conduct a credibility assessment and comprehensively evaluate the credibility of the results. ; ; in For instrument accuracy reliability (based on instrument calibration status). Reliability of environmental stability (based on the amplitude of environmental fluctuations). For statistical significance confidence (based on p-value). For repeatability confidence (based on the coefficient of variation of repeated experiments), , , and These represent the corresponding credibility weight coefficients; S54. Perform a consistency check, comparing the results with known patterns. and historical data Perform a consistency check ; ; in It serves as a distance metric between the results and known patterns. S55. Generate a structured verification report containing verification conclusions, confidence levels, and evidence based on the multidimensional verification results.

[0030] S6. Based on the verification report, check whether there are any experimental failures or results that do not meet expectations. If so, perform multidimensional failure attribution; otherwise, proceed to S7. The process of performing multidimensional failure attribution is as follows: S61. Perform hypothesis attribution, determine whether the scientific hypothesis itself is flawed, and calculate the degree of contradiction between the hypothesis and the results. The process is as follows: ; When the degree of contradiction exceeds the preset contradiction threshold When the hypothesis is established, determine that the attribution is valid and analyze the direction in which the hypothesis needs to be revised. S62. Perform execution attribution to determine if there were any operational deviations during the execution process; and backtrack the execution records to calculate the execution deviation index. ; ; in and The first Actual and planned parameters for each step of the operation; when When the attribution for execution is established, This indicates the preset execution deviation threshold; S63. Perform instrument attribution to determine if there is an instrument malfunction or insufficient accuracy; check instrument status records and calibration data, and calculate the contribution of instrument deviation. : ; in This is due to instrument system deviation. For random error, Confidence coefficient; when When the instrument attribution is established, This indicates the preset instrument deviation threshold. S64. Perform environmental attribution, determine whether environmental conditions exceed expectations, and calculate the environmental deviation index. : ; in Indicates actual environmental conditions. Indicates expected environmental conditions. Represents a set of time points; when When the environmental attribution is deemed valid, This indicates the preset environmental deviation threshold; S65. Perform reagent attribution to determine if there are any reagent quality issues; check reagent batches, storage conditions, and expiration dates, and calculate reagent quality fractions. : ; in The passing score for the batch. For storage condition compliance score, For the validity period compliance score, , and These represent the corresponding compliance score weighting coefficients; when When the attribution of the reagent is deemed valid, This indicates the preset reagent quality threshold; S66. Perform attribution synthesis judgment, calculate the attribution confidence of each dimension, and generate attribution vectors. ; ; in These represent the confidence levels of the corresponding dimension attributions in S61 to S66, respectively. Attribution confidence is calculated using a weighted average of the strength of evidence: ; S67. Generate a structured attribution report, including attribution conclusions, chain of evidence, and improvement recommendations; S68. Based on the attribution results, generate adjustment suggestions for the next round of experiments.

[0031] ; in To adjust the suggestion generation function, the output includes the direction of hypothesis revision, the implementation of improvement measures, and the condition adjustment plan; S7. Perform model write-back and iterative evolution.

[0032] S71. Update the property parameters and dynamic model of the experimental object; ; in This represents the updated reaction kinetic parameters. This represents the current reaction kinetic parameters. This indicates the learning rate for updating the dynamic parameters. The loss function representing the dynamic model. Indicates the model's predicted value; S72. Update the instrument's accuracy model and reliability model; S73. Accumulate experimental experience and update knowledge. ; S74, Parameters of the optimization scheme generation strategy; S75. Update the threshold of the risk prediction model; S76. Generate an experimental evolution log to record the inputs, outputs, conclusions, and improvements for each round of experiments; S77. Return to step S2 and proceed to the next experimental cycle.

[0033] The unmanned laboratory scheme generation and multidimensional failure attribution method are implemented through a corresponding system; the system includes a candidate experimental scheme generation module for generating candidate schemes through three strategies: hypothesis-driven, analogical reasoning, and exploratory. Constraint filtering and ranking module used for feasibility filtering and comprehensive ranking of candidate solutions; An execution flow orchestration module used for task decomposition, dependency analysis, parallel optimization, and resource scheduling of selected solutions; An experimental execution and real-time monitoring module used to execute experimental procedures and monitor process status; A multi-dimensional verification module for performing four-dimensional verification of experimental results: target conformity, hypothesis testing, credibility, and consistency. A multidimensional failure attribution module for attributing experimental failures across five dimensions: hypothesis, execution, instrument, environment, and reagents. The Model Writeback and Loop Evolution module is used to write back experimental results and attribution analysis to the experimental world model.

[0034] Taking the high-entropy alloy hardness optimization experiment as an example, the specific implementation process of the method of the present invention is explained as follows: Implementation of step S1: Construct an experimental world model. Target layer: The research objective is "to increase the hardness to above HV 400 in CoCrFeNi-based high-entropy alloys by adding Al elements"; the scientific hypothesis is "the Al content in the range of 5-15 at.% is positively correlated with hardness"; the expected result is "HV ≥ 400"; the success criterion is "hardness measurement value ≥ 400 HV and repeatability standard deviation < 10 HV".

[0035] The experimental material was a CoCrFeNi-based high-entropy alloy. The sample properties included grain size, phase composition, and density. The reaction kinetic parameters included solidification rate and phase transformation temperature.

[0036] Instrumentation layer: Available instruments include an electric arc melting furnace (available, accuracy ±2℃), a hardness tester (available, accuracy ±5HV), an XRD diffractometer (available), and a SEM microscope (under maintenance, available in 3 days).

[0037] Environmental layer: laboratory temperature 22±1℃, humidity 45±5%, cleanliness level ISO 7.

[0038] Knowledge layer: The domain knowledge graph includes high-entropy alloy phase diagrams and literature data on known Al content-hardness relationships; safety constraints include the maximum temperature limit for arc melting and inert gas protection requirements.

[0039] Researcher intent layer: Researchers prefer conservative allocation strategies, have low risk tolerance, and prioritize safety.

[0040] Implementation of step S2: Hypothesis-driven generation: Based on the hypothesis that "Al content is positively correlated with hardness", a gradient experiment was designed with Al contents of 5, 8, 10, 12 and 15 at.%, and the control variables were melting temperature (1800℃) and cooling rate (10℃ / s).

[0041] Analogical reasoning generation: Retrieve similar experiments (hardness experiments of FeCoNiCrAl system) in the knowledge layer, extract successful solutions (the highest hardness is achieved when the Al content is 12 at.%), and generate analogy solutions.

[0042] Exploratory generation: Based on Bayesian optimization, three exploratory schemes (knowledge gradient acquisition functions) are generated in a two-dimensional parameter space with an Al content of 5-15 at.% and a melting temperature of 1700-1900℃.

[0043] Constraint filtering: SEM microscope is unavailable (under maintenance), filtering schemes requiring SEM characterization; Arc melting temperature limit is 1900℃, filtering schemes exceeding the temperature limit. Six feasible schemes remain after filtering.

[0044] Scheme ranking: Calculate the overall score. Scheme "Al=12 at.%, temperature 1800℃" scored the highest (high information gain, low cost, low risk, and matches the conservative preferences of researchers), and was selected as the implementation scheme.

[0045] Implementation of step S3: Task breakdown: The solution is broken down into 8 atomic operations: batching → mixing → electric arc melting → remelting → casting → cutting → grinding → hardness testing.

[0046] Dependency analysis: The process of batching → mixing → arc melting is time-dependent; arc melting requires the use of inert gas (resource dependency); hardness testing must be performed after grinding is completed (time-dependent).

[0047] Parallel optimization: XRD testing can be performed in parallel with hardness testing (using different instruments), reducing the total execution time by 2 hours.

[0048] Risk assessment: The historical failure rate of the arc melting step is 5% (electrode contamination), and the historical failure rate of the hardness testing step is 2% (uneven sample surface).

[0049] Implementation of step S4: The experiment was conducted. During the arc melting step, the process monitoring detected a temperature deviation of Δ=8℃ in the molten pool (threshold 5℃), triggering an anomaly alarm. This was determined to be temperature sensor drift (anomaly type: instrument deviation, severity: low). The process was paused for 30 seconds to recalibrate the sensor before resuming execution.

[0050] Implementation of step S5: Experimental results: The sample hardness of Al=12 at.% was HV 412±6 (3 ​​replicates).

[0051] Target compliance verification: Δ_goal = |412-400| / 400 = 3%<5% (threshold), the target is deemed to be compliant.

[0052] Hypothesis verification: t_stat = (412-380) / (15 / √3) = 3.7>t_{0.025,2} = 4.303, at the α=0.05 level, we do not reject the null hypothesis (Al content is positively correlated with hardness).

[0053] Confidence assessment: C_inst = 0.95 (hardness tester is well calibrated), C_env = 0.98 (environmentally stable), C_stat = 0.92 (p=0.03), C_repeat = 0.96 (coefficient of variation 1.5%). Overall confidence level: Conf = 0.95.

[0054] Implementation of step S6: The experiment was successful; there's no need to attribute failures. Record the successful experiences.

[0055] Implementation of step S7: Update the experimental world model: add "Al=12 at.%, HV=412" to the knowledge layer; update the Al content-hardness relationship model; update the researcher intent layer (confirming the effectiveness of the conservative strategy).

[0056] Example 2: Attribution of failures in chemical experiments; Taking the experiment of low yield of catalytic reaction as an example, the implementation process of multidimensional failure attribution is explained.

[0057] Experimental objective: To hydrogenate substrate A to product B using a Pd / C catalyst, with an expected yield ≥90%. Actual result: The yield was only 62%.

[0058] Five-dimensional attribution analysis: Hypothesis attribution: (Yield ≥ 90%) Hypothesis attribution confidence level: 0.35. Analysis: The hypothesis "yield ≥ 90% under standard conditions" may not be applicable to the current substrate.

[0059] Execution attribution: Backtracking the execution log revealed that the actual hydrogen flow rate was 80 mL / min (planned 100 mL / min), and ExecDev = 0.20 > θ_exec = 0.10. Execution attribution confidence level: 0.40. Analysis: Insufficient hydrogen flow rate led to incomplete reaction.

[0060] Instrument Attribution: Inspection of the mass flow meter calibration record revealed that the flow meter was last calibrated 3 months ago (the recommended cycle is 1 month), with InstDev = 0.15 < θ_inst = 0.20. Instrument attribution confidence level: 0.10. Analysis: The flow meter may have a deviation, but it has not reached the attribution threshold.

[0061] Environmental attribution: Environmental monitoring data showed that the temperature fluctuated by ±0.5℃ during the experiment (within the normal range), EnvDev = 0.03 < θ_env = 0.10. Confidence level of environmental attribution: 0.05.

[0062] Reagent attribution: A review of the Pd / C catalyst batch records revealed that this batch of catalyst had been opened for more than 30 days (use is recommended within 15 days of opening). Q_reagent = 0.65 < θ_reagent = 0.80. Reagent attribution confidence level: 0.10. Analysis: The catalyst may have been partially deactivated due to exposure to air.

[0063] Attribution analysis: The attribution vector is A = [0.35, 0.40, 0.10, 0.05, 0.10]. The primary attribution is performance attribution (insufficient hydrogen flow, confidence level 0.40), and the secondary attribution is hypothesis attribution (the hypothesis may not apply to the current substrate, confidence level 0.35).

[0064] Recommended adjustments for the next round: (1) Re-test after calibrating the hydrogen mass flow meter; (2) Increase the hydrogen flow rate to 120 mL / min to compensate for possible flow deviations; (3) Replace with a new batch of Pd / C catalyst; (4) If the yield is still below 80%, revise the assumption to "the yield of the substrate is 70-85% under standard conditions".

[0065] Example 3: Environmental attribution in biological experiments; Taking cell culture experiments as an example. Experimental goal: To culture HeLa cells at 37℃ and 5% CO2 for 48 hours, with an expected cell viability of ≥95%. Actual result: The viability was only 78%.

[0066] Attribution analysis revealed that environmental monitoring data showed the CO2 concentration in the incubator decreased to 3.2% (target 5%) between hours 12 and 18 of the experiment, with EnvDev = 0.45 > θ_env = 0.10. The environmental attribution confidence level was 0.75. Further investigation revealed that the CO2 cylinder was depleted during the experiment.

[0067] Recommended adjustments for the next round: (1) Replace the CO2 cylinder and add an automatic switching function for the spare cylinder; (2) Re-execute the experiment; (3) Add a CO2 balance pre-check step to the experimental procedure.

[0068] Therefore, the present invention employs a world model-based unmanned laboratory scheme generation and multidimensional failure attribution method, which has the following beneficial effects: (1) Through the six-layer experimental world model, a unified expression of all experimental elements is realized, enabling the experimental system to have the ability to "understand the experiment"; (2) By flexibly combining three scheme generation strategies (hypothesis-driven, analogical reasoning, and exploratory), a systematic scheme generation was achieved, covering both verification experiments and exploratory experiments. (3) By using constraint filtering and execution process orchestration, ensure the feasibility of the solution during the execution phase and avoid the problem of "design is feasible but execution fails"; (4) Verify the results in four dimensions (target compliance, hypothesis testing, credibility, and consistency) to ensure the reliability of the experimental conclusions; (5) By using five-dimensional failure attribution (hypothesis, execution, instrument, environment, reagent), the causes of failure can be systematically identified, providing clear directions for improvement in the next round of experiments; (6) Through model rewriting and iterative evolution, we can achieve continuous accumulation of experimental knowledge and continuous improvement of experimental capabilities.

[0069] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for generating unmanned laboratory solutions and attributing failures in a multidimensional manner based on a world model, characterized in that, Includes the following steps: S1. Construct an experimental world model; S2. Based on the experimental world model, generate candidate experimental schemes, perform constraint filtering and comprehensive ranking on the candidate schemes, and select the optimal scheme. S3. Arrange the execution flow for the selected optimal solution; S4. Conduct the experimental procedure according to the programmed results, and monitor it in real time. S5. Perform multidimensional verification of the experimental results and generate a verification report; S6. Based on the verification report, check whether there are any experimental failures or results that do not meet expectations. If so, perform multidimensional failure attribution; otherwise, proceed to S7. S7. Perform model write-back and iterative evolution.

2. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 1, characterized in that: In S1, the experimental world model is obtained by constructing the target layer, object layer, instrument layer, environment layer, knowledge layer, and researcher intent layer. ; S11. Construct the target layer Including research objectives Scientific Hypotheses Expected Results Successful judgment criteria and priority ; S12, Constructing the object layer Including the state of experimental materials Sample properties Reaction kinetic parameters and historical experimental data ; S13, Constructing the Instrument Layer Including instrument availability Accuracy status Calibration status Operational constraints and maintenance plan ; Define the instrument accuracy model The process is as follows: ； in For instruments The systematic bias, For instruments The standard deviation of random error This represents the true value of the measurement. Indicator Instrument The random error term; S14, Constructing the Environment Layer Including temperature ,humidity Cleanliness ,vibration and electromagnetic environment ; Define the environment state vector The process is as follows: ； S15, Building the Knowledge Layer Including domain knowledge graphs Known patterns Similar experimental experience and safety constraints ; S16. Constructing the Researcher Intent Layer Including researchers' goal statements Constraint preferences Risk tolerance and historical revision records ; S17. Establish an experimental world model based on the six levels constructed in S11-S16. The process is as follows: 。 3. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 2, characterized in that, The process of S2 is as follows: S21. Conduct hypothesis-driven generation based on current scientific hypotheses. Derive a verifiable experimental design; identify the independent variables. Dependent variable and control variables At the same time, the solution must meet the requirement of identifiability; S22. Perform analogical reasoning to generate successful experiences of similar experiments in the knowledge layer, and define the experimental similarity. The process is as follows: ； in For similar experiments and Target similarity, For similar experiments and The similarity of the experimental subjects For similar experiments and Similarity of experimental conditions These are the corresponding similarity weights; Select the one with the highest similarity We extracted successful solutions from historical experiments as analogies and adapted them to the current experimental conditions. S23. Perform exploratory generation: Generate exploratory solutions in the parameter space based on Bayesian optimization, with the goal of maximizing information gain. The process is as follows: ； in Indicates the data acquisition function. This indicates the parameter values ​​for the next exploratory experiment. Represents historical data; S24. Generate a set of candidate experimental schemes based on the three strategies in S21-S23. ; S25. On the set of candidate solutions Perform constraint filtering to filter out infeasible solutions. The process is as follows: ； in Indicates a feasible solution. For instrument availability checks, For reagent compatibility testing, For safety constraint checks, For environmental condition inspection; S26. Rank the feasible solutions comprehensively and define the ranking criteria. : ； in For the plan The expected information gain, For execution costs, For risk level, To ensure a good match with the researchers' intentions, , , and These are the weighting coefficients for expected information gain, execution cost, risk level, and the degree of matching of researchers' intentions, respectively. S27. Based on the ranking results, find the solution with the highest score, i.e., the optimal solution. ; 。 4. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 3, characterized in that: The acquisition function in S23 can be either the expected improvement function or the knowledge gradient function.

5. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 4, characterized in that, The process of S3 is as follows: S31. Decompose the task into a sequence of atomic operations. Each atomic operation Defined as a quintuple; ； in For operation type, For operation parameters, For the expected duration, For the required resources, As a prerequisite; S32. Perform dependency analysis and construct an operation dependency graph. , among which the side Indicates operation Dependent on operation The completion; Dependency types include time-series dependencies, resource dependencies, and data dependencies; S33. Perform parallel optimization: based on the dependency graph, perform topological sorting to identify the set of operations that can be executed in parallel, and calculate the shortest execution time using the critical path method. : ； in Indicates the expected duration; S34. Conduct risk assessment: assess the risk of each operational step and calculate the probability of failure. The process is as follows: ； in This represents the historical failure rate for this type of operation. For operational complexity, The current health status of the instruments used. This represents the failure probability bias term. , and These represent the weighting coefficients corresponding to historical failure rate, operational complexity, and the current health status of the instrument used; S35. Develop contingency plans for situations where the probability of failure exceeds a preset failure threshold. The key steps include developing contingency plans and rollback strategies for failures. S36. Perform resource scheduling, coordinate instrument usage time, reagent consumption, and environmental control requirements, and avoid resource conflicts through constraint satisfaction problem solving.

6. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 5, characterized in that, The process of S4 is as follows: S41. During execution, process monitoring is performed, and each operation step is monitored. Monitor the deviation between its actual execution status and the expected status. ; ； in Indicates the actual execution status. Indicates the expected state; when An abnormal alarm is triggered at any time. This indicates the preset abnormal threshold. S42. Anomaly detection based on statistical process control methods The process is as follows: ； in and These are the mean and standard deviation of the process variable, respectively. S43. Make a decision on the handling of the situation, and decide on the corresponding handling of the anomaly, such as continuing execution, pausing and waiting, adjusting parameters, or terminating the experiment, based on the type and severity of the anomaly. S44. Completely record all operations, observations, and decisions during the experiment to form a traceable experimental log.

7. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 6, characterized in that, The process of performing multidimensional verification of experimental results and generating a verification report in S5 is as follows: S51. Conduct target conformity verification and calculate the deviation between the experimental results and the expected target. The process is as follows: ； in Indicates the experimental results, Indicates the expected goal; when The target is determined to be compatible at that time. This indicates the preset deviation threshold. S52. Perform hypothesis testing. Determine whether the results support or refute the original hypothesis through hypothesis testing, and calculate the test statistic. : ； in This represents the mean of the experimental results. The expected value under the assumption, The standard deviation of the sample is 1. This represents the number of times the experiment was repeated. when At the significance level We reject the null hypothesis. This indicates a preset statistical threshold. S53. Conduct a credibility assessment and comprehensively evaluate the credibility of the results. ; ； in To ensure the accuracy and reliability of the instrument, For environmental stability and credibility, To assess statistical significance and reliability, For repeatability reliability, , , and These represent the corresponding credibility weight coefficients; S54. Perform a consistency check, comparing the results with known patterns. and historical data Perform a consistency check ; ； in It serves as a distance metric between the results and known patterns. S55. Generate a structured verification report containing verification conclusions, confidence levels, and evidence based on the multidimensional verification results.

8. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model according to claim 7, characterized in that, The process of multidimensional failure attribution in S6 is as follows: S61. Perform hypothesis attribution, determine whether the scientific hypothesis itself is flawed, and calculate the degree of contradiction between the hypothesis and the results. The process is as follows: ； When the degree of contradiction exceeds the preset contradiction threshold When the hypothesis is established, determine that the attribution is valid and analyze the direction in which the hypothesis needs to be revised. S62. Perform execution attribution to determine whether there are operational deviations during the execution process; It also backtracks the execution records and calculates the execution deviation index. ; ； in and The first Actual and planned parameters for each step of the operation; when When the attribution for execution is established, This indicates the preset execution deviation threshold; S63. Perform instrument attribution to determine if there is an instrument malfunction or insufficient accuracy; check instrument status records and calibration data, and calculate the contribution of instrument deviation. : ； in This is due to instrument system deviation. For random error, Confidence coefficient; when When the instrument attribution is established, This indicates the preset instrument deviation threshold. S64. Perform environmental attribution, determine whether environmental conditions exceed expectations, and calculate the environmental deviation index. : ； in Indicates actual environmental conditions. Indicates expected environmental conditions. Represents a set of time points; when When the environmental attribution is deemed valid, This indicates the preset environmental deviation threshold; S65. Perform reagent attribution to determine if there are any reagent quality issues; Check the reagent batch, storage conditions, and expiration date; calculate the reagent mass fraction. : ； in The passing score for the batch. For storage condition compliance score, For the validity period compliance score, , and These represent the corresponding compliance score weighting coefficients; when When the attribution of the reagent is deemed valid, This indicates the preset reagent quality threshold; S66. Perform attribution synthesis judgment, calculate the attribution confidence of each dimension, and generate attribution vectors. ; ； in These represent the confidence levels of the corresponding dimension attributions in S61 to S66, respectively. S67. Generate a structured attribution report, including attribution conclusions, chain of evidence, and improvement recommendations; S68. Based on the attribution results, generate adjustment suggestions for the next round of experiments.

9. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model as described in claim 8, characterized in that, The process for S7 is as follows: S71. Update the property parameters and dynamic model of the experimental object; ； S72. Update the instrument's accuracy model and reliability model; S73. Accumulate experimental experience and update knowledge. \mathcal{L}_{\text{know}} ; S74, Parameters of the optimization scheme generation strategy; S75. Update the threshold of the risk prediction model; S76. Generate an experimental evolution log to record the inputs, outputs, conclusions, and improvements for each round of experiments; S77. Return to step S2 and proceed to the next experimental cycle.

10. The method for generating unmanned laboratory schemes and attributing multidimensional failures based on a world model according to claim 9, characterized in that: The unmanned laboratory scheme generation and multidimensional failure attribution method are implemented through a corresponding system; the system includes a candidate experimental scheme generation module for generating candidate schemes through three strategies: hypothesis-driven, analogical reasoning, and exploratory. Constraint filtering and ranking module used for feasibility filtering and comprehensive ranking of candidate solutions; An execution flow orchestration module used for task decomposition, dependency analysis, parallel optimization, and resource scheduling of selected solutions; An experimental execution and real-time monitoring module used to execute experimental procedures and monitor process status; A multi-dimensional verification module for performing four-dimensional verification of experimental results: target conformity, hypothesis testing, credibility, and consistency. A multidimensional failure attribution module for attributing experimental failures across five dimensions: hypothesis, execution, instrument, environment, and reagents. The Model Writeback and Loop Evolution module is used to write back experimental results and attribution analysis to the experimental world model.

Images