AI failure mode dynamic diagnosis method and system based on sequence mode and function data analysis

Abstract

The embodiment of the invention discloses an AI failure mode dynamic diagnosis method and system based on a sequence mode and function data analysis, which realize the crossing of AI interaction failure from static type identification to dynamic process diagnosis and improve the depth and timeliness of failure analysis. Accurate and explainable data support is provided for optimization of an artificial intelligence system. Through a sequence pattern mining and function data analysis technology, a failure pattern generated in an interaction process of a user and an artificial intelligence model (such as a large language model, a generative AI or an AI intelligent agent) is identified, analyzed and dynamically tracked, and the influence of the failure pattern is analyzed.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, specifically to a dynamic diagnosis method and system for AI failure modes based on sequence pattern and function data analysis. Background Technology

[0002] With artificial intelligence (AI) technology widely applied in fields such as conversational systems, intelligent customer service, and content generation, interaction between AI systems and users has become an important mode of human-machine collaboration. However, AI often exhibits interaction failures during conversations with users, such as giving irrelevant answers, misunderstandings, and inappropriate emotional responses. These failures directly impact user experience, satisfaction, and willingness to continue using the system, and can even lead to complaints and trust crises.

[0003] Currently, analytical methods for AI interaction failures mainly focus on single technical approaches. For example, some studies use sequence pattern mining algorithms (such as Seq2Pat) to extract frequently occurring failure patterns from conversation history, but these are often limited to specific scenarios (such as emotion recognition or cognitive language analysis) and lack the ability to comprehensively identify multiple types of failures (such as emotional, cognitive, and ethical failures). Other studies use functional data analysis methods to process conversation data, analyzing the distribution of specific failure types over time by modeling changes in language features along a continuous time dimension, but these methods fail to be effectively combined with sequence pattern mining, thus failing to achieve a holistic diagnosis from pattern recognition to temporal dynamic analysis.

[0004] The limitations of existing technologies are mainly reflected in:

[0005] Narrow scope of identification: Existing methods mostly analyze a certain type of failure, which makes it difficult to cover the multiple failure types that may occur in AI interaction;

[0006] Lack of dynamic perspective: Most methods treat failure as discrete events and fail to analyze its occurrence patterns, evolution trends and temporal correlations between types on a continuous time axis;

[0007] Disconnect between diagnosis and response: Failure to effectively combine failure identification with impact assessment makes it difficult for enterprises to adjust artificial intelligence systems in a timely and accurate manner. Summary of the Invention

[0008] To address this, embodiments of the present invention provide a dynamic diagnostic method and system for AI failure modes based on sequence pattern and function data analysis, in order to solve the technical problems of existing technologies that focus on a single technical path, resulting in a narrow recognition range and a lack of dynamic perspective in diagnosis and response.

[0009] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0010] According to a first aspect of the present invention, a dynamic diagnosis method for AI failure modes based on sequence pattern and function data analysis is provided, the method comprising:

[0011] The system acquires raw conversation sequence data generated from multiple rounds of interaction between the user and the artificial intelligence system, and cleans and structures the raw conversation sequence data to form a standardized interactive conversation dataset.

[0012] Based on sequence pattern mining technology, the standardized interactive session dataset is analyzed, and frequently occurring session patterns that meet preset failure criteria are extracted as candidate failure patterns.

[0013] The candidate failure patterns are matched and classified with a predefined failure type knowledge base. The failure types include at least emotional indifference failure, cognitive breakdown failure, and ethical anomie failure. The classified failure patterns are then manually or automatically verified.

[0014] Using functional data analysis methods, an occurrence intensity function is constructed for each verified failure mode on a continuous interaction time axis. The occurrence intensity function is used to characterize the probability or frequency of occurrence of the failure mode at different stages of the session process.

[0015] The relationship between the occurrence intensity functions of different failure modes on the time axis is analyzed, including temporal sequence relationship, co-occurrence relationship or mutual attrition relationship, and a dynamic diagnostic report on the interaction failure characteristics of the artificial intelligence system is generated based on the relationship.

[0016] Furthermore, based on sequence pattern mining technology, the standardized interactive session dataset is analyzed to extract frequently occurring session patterns that meet preset failure criteria as candidate failure patterns, including:

[0017] The sequence pattern mining technology employs the Seq2Pat algorithm, the PrefixSpan algorithm, or a sequence pattern learning model based on a recurrent neural network; the preset failure criteria include a pattern support threshold, a confidence threshold, and a failure keyword matching rule based on semantic similarity.

[0018] Furthermore, in the failure type knowledge base, each failure type is further subdivided into at least two subcategories; among them, the subcategories of emotional indifference failure include lack of empathy, inappropriate emotional response, and misuse of emotional vocabulary; the subcategories of cognitive dissonance failure include factual errors, logical contradictions, biased understanding of context, and omission of instruction execution; and the subcategories of ethical misconduct failure include biased responses, leakage of privacy information, and generation of harmful content.

[0019] Furthermore, using functional data analysis methods, an occurrence intensity function on a continuous interaction timeline is constructed for each verified failure mode, including:

[0020] Normalize the timeline of the entire interactive session;

[0021] Map each failure mode instance to a corresponding time point on the normalized time axis;

[0022] By employing nonparametric smoothing techniques or basis function expansion, the distribution of discrete failure mode instances on the time axis is fitted to generate a continuous occurrence intensity curve, i.e., the occurrence intensity function.

[0023] Furthermore, the content of the dynamic diagnostic report includes:

[0024] A list of core failure modes and their respective types;

[0025] Intensity trends of various failure modes in the early, middle, and late stages of interaction;

[0026] A critical failure mode cluster, which consists of multiple failure modes that are highly correlated or frequently co-occur on the time axis;

[0027] Recommendations for optimizing AI systems based on time-dynamic features.

[0028] Furthermore, the artificial intelligence system includes, but is not limited to, generative large-scale language models, task-oriented conversational robots, emotional companionship AI, or intelligent customer service systems; the original conversation sequence data is in text form, or text converted from voice conversations.

[0029] Furthermore, the automated verification is achieved by training a binary classification verification model, which takes the candidate failure pattern and its context as input and outputs a confidence score that the pattern is judged as a true failure.

[0030] Furthermore, the method also includes:

[0031] The generated dynamic diagnostic report is correlated with user satisfaction scores, session termination rates, or user complaint data collected concurrently to quantitatively analyze the substantial impact of the dynamic characteristics of specific failure modes on user interaction experience, and the analysis results are fed back to the training or optimization module of the artificial intelligence system.

[0032] According to a second aspect of the present invention, an AI failure mode dynamic diagnosis system based on sequence pattern and function data analysis is provided, the system comprising:

[0033] The data preprocessing module is used to acquire the raw conversation sequence data generated by the multi-round interaction between the user and the artificial intelligence system, and to clean and structure the raw conversation sequence data to form a standardized interactive conversation dataset.

[0034] The sequence pattern mining module is used to analyze the standardized interactive session dataset based on sequence pattern mining technology, and extract frequently occurring session patterns that meet preset failure criteria as candidate failure patterns.

[0035] The pattern classification and verification module is used to match and classify the candidate failure patterns with a predefined failure type knowledge base. The failure types include at least emotional indifference failure, cognitive breakdown failure, and ethical anomie failure. The module also performs manual or automated verification on the classified failure patterns.

[0036] The function data analysis module is used to construct an occurrence intensity function on a continuous interaction time axis for each type of verified failure mode using function data analysis methods. The occurrence intensity function is used to characterize the probability or frequency of occurrence of the failure mode at different stages of the session process.

[0037] The dynamic diagnostic report generation module is used to analyze the relationship between the occurrence intensity functions of different failure modes on the time axis, including temporal sequence relationship, co-occurrence relationship or attrition relationship, and generate a dynamic diagnostic report on the interaction failure characteristics of the artificial intelligence system based on the relationship.

[0038] Furthermore, the dynamic diagnostic report generation module integrates an association rule mining unit, which is used to discover temporal association rules between different failure mode intensity functions, and predict other failure modes that may be triggered in a specific interaction stage based on the rules.

[0039] The embodiments of the present invention have the following advantages:

[0040] This invention first acquires and preprocesses user interaction session data with an artificial intelligence system. Second, it uses a sequence pattern mining algorithm to extract high-frequency candidate failure patterns and classifies them into specific types, such as emotional indifference failure and cognitive discontinuity failure. Then, through function data analysis, discrete failure events are fitted into intensity functions on a continuous normalized time axis. Function features are then extracted and principal component analysis is performed to characterize the dynamic evolution of failure patterns. Finally, the correlation between different failure patterns on the time axis and their impact on user behavior are analyzed to generate a dynamic diagnostic report. This invention achieves a leap from static type identification to dynamic process diagnosis of AI interaction failures, improving the depth and timeliness of failure analysis and providing accurate and interpretable data support for the optimization of artificial intelligence systems. Through sequence pattern mining and function data analysis techniques, failure patterns generated during user interaction with artificial intelligence models (such as large language models, generative AI, or AI agents) are identified, analyzed, and dynamically tracked, and their impact is analyzed. Attached Figure Description

[0041] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0042] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0043] Figure 1 A schematic diagram of the logical structure of an AI failure mode dynamic diagnosis system based on sequence pattern and function data analysis provided in an embodiment of the present invention;

[0044] Figure 2 This is a flowchart illustrating a dynamic diagnostic method for AI failure modes based on sequence pattern and function data analysis, provided in an embodiment of the present invention. Detailed Implementation

[0045] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0046] AI interaction failure refers to an event in which the AI ​​fails to answer the question, leading to user dissatisfaction during the user-AI interaction session, i.e., the round-trip interaction process of "user-question, AI-answer".

[0047] Algorithms applied to conversation data processing focus only on specific conversation scenarios, such as conversations involving "user emotions and AI attitudes," or on recognizing emotional language and cognitive language, without simultaneously considering algorithms that can identify multiple potential scenarios and have multiple possible failures.

[0048] Currently, in session data processing, sequence pattern mining algorithms only include the Seq2Pat algorithm, and it seems there's no use of deep learning or reinforcement learning to learn failure types. Function data analysis, in processing session data, can pre-define when specific failure types are likely to occur in consecutive time points by backtracking from performance data. However, there's no combined sequence pattern mining algorithm and function data analysis that can both identify the type of failure and analyze this discrete failure type on a timeline.

[0049] To address the technical problems of narrow recognition range and lack of dynamic perspective diagnosis and response caused by focusing on a single technical path.

[0050] refer to Figure 1 This invention discloses an AI failure mode dynamic diagnosis system based on sequence pattern and function data analysis. The system includes: a data preprocessing module 1; a sequence pattern mining module 2; a pattern classification and verification module 3; a function data analysis module 4; and a dynamic diagnosis report generation module 5.

[0051] Corresponding to the AI ​​failure mode dynamic diagnosis system based on sequence pattern and function data analysis disclosed above, this invention also discloses an AI failure mode dynamic diagnosis method based on sequence pattern and function data analysis. The following details the AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis disclosed in this invention, in conjunction with the AI ​​failure mode dynamic diagnosis system based on sequence pattern and function data analysis described above.

[0052] refer to Figure 2This invention discloses a dynamic diagnosis method for AI failure modes based on sequence pattern and function data analysis. It obtains the original conversation sequence data generated by multiple rounds of interaction between the user and the artificial intelligence system, and cleans and structures the original conversation sequence data to form a standardized interactive conversation dataset.

[0053] Based on sequence pattern mining technology, the standardized interactive session dataset is analyzed, and frequently occurring session patterns that meet preset failure criteria are extracted as candidate failure patterns.

[0054] The candidate failure patterns are matched and classified with a predefined failure type knowledge base. The failure types include at least emotional indifference failure, cognitive breakdown failure, and ethical anomie failure. The classified failure patterns are then manually or automatically verified.

[0055] Using functional data analysis methods, an occurrence intensity function is constructed for each verified failure mode on a continuous interaction time axis. The occurrence intensity function is used to characterize the probability or frequency of occurrence of the failure mode at different stages of the session process.

[0056] The relationship between the occurrence intensity functions of different failure modes on the time axis is analyzed, including temporal sequence relationship, co-occurrence relationship or mutual attrition relationship, and a dynamic diagnostic report on the interaction failure characteristics of the artificial intelligence system is generated based on the relationship.

[0057] Data acquisition and preprocessing:

[0058] This step aims to prepare a structured functional data foundation for subsequent analysis.

[0059] Raw data collection: Collect complete user-AI conversation data from the logs of the target AI system (such as a customer service robot or a generative conversational assistant). Each conversation contains multi-turn interaction records sorted by timestamps.

[0060] Session serialization and annotation:

[0061] Each session undergoes basic cleaning processes such as word segmentation and noise reduction.

[0062] A natural language processing model is used to annotate the user utterances and AI responses in each round of conversation with functional features.

[0063] Language function categories: Reference method, annotation such as cognitive language (e.g., "believe", "understand"), affective language (e.g., "happy", "disappointed"), social language (e.g., "we", "cooperate"), deterministic language (e.g., "certain", "inevitable"), etc.

[0064] Conversational behavior types: such as asking questions, answering questions, complaining, clarifying, and providing information.

[0065] Failure risk label: Based on predefined rules or a simple classifier, an initial failure risk score (e.g., 0-1 points) is given for each round of AI response.

[0066] Constructing initial functional data: Each session is treated as an observation unit. Its labeled features (such as "affective language intensity" and "failure risk value") form observations at discrete session turn times. These discrete point sequences serve as the raw input for subsequent functional data analysis, with the goal of transforming them into a smooth, continuous function.

[0067] Interaction failure pattern identification and classification:

[0068] The core of this step is to use sequence pattern mining technology to locate and define specific failure modes from discrete conversation behavior / feature sequences.

[0069] Sequence pattern mining:

[0070] The input is a richly annotated sequence of conversational behaviors obtained in the previous step.

[0071] Algorithms such as Seq2Pat and PrefixSpan are used to mine frequently occurring subsequence patterns that span multiple rounds of conversation. For example, a high-frequency pattern P is: [User-Question (High Cognition) -> AI-Answer (Low Certainty) -> User-Complaint (Negative Emotion)].

[0072] By combining domain knowledge with constraints (such as the pattern must include "user complaints" or "low certainty of AI response"), candidate failure patterns that are highly correlated with interaction failures are selected.

[0073] Failure Mode Definitions and Classifications:

[0074] Each candidate failure mode is assigned a specific failure type definition, for example:

[0075] Cognitive Disruption Failure: This pattern manifests as logical contradictions or omissions of key information in the AI's response after the user poses a complex, multi-constrained question.

[0076] Emotionally indifferent failure: This pattern is characterized by the AI ​​responding with neutral, formulaic, or emotionally inconsistent content after the user expresses strong emotions (positive / negative).

[0077] Ethical anomie type of failure: The pattern is characterized by biased language, inappropriate advice, or privacy risks in AI responses.

[0078] This step outputs a discrete set of failure events, each associated with a specific session, the specific round (time point) in which it occurred, and its failure type.

[0079] Failure dynamic modeling and analysis based on function data analysis:

[0080] Time axis normalization and functionalization:

[0081] Map the absolute timeline of the session (e.g., from round 1 to round N) to the standard interval [0, 1], where 0 indicates the start of the session and 1 indicates the end of the session.

[0082] Function smoothing: For a certain type of failure (such as "cognitive breakdown"), it may occur multiple times in a session (discrete points). Using B-spline basis function expansion or kernel smoothing, the discrete time points of all intermediate failure events in the session are fitted to a continuous intensity function f(t) defined on [0,1]. The value of f(t) represents the probability density or intensity of this type of failure near the normalized time t. Function feature extraction and deep analysis (integrating literature methods):

[0083] Dynamic indicator construction: Drawing on the analysis of "price level, speed, and acceleration" in the literature, a more refined calculus analysis is performed on the failure intensity function f(t):

[0084] f(t): Failure intensity (level).

[0085] f'(t): The first derivative of the failure intensity, reflecting the rate of change of failure risk. A positive value indicates that the risk is accumulating and accelerating, while a negative value indicates that it is mitigating.

[0086] f''(t): The second derivative of the failure intensity, reflecting the acceleration of the rate of change, can be used to identify the inflection point of intensity change (such as the turning point from accelerated deterioration to the beginning of relief).

[0087] Functional Principal Component Analysis (FPCA): As described in the literature, FPCA is performed on a set of failure intensity functions {f_i(t)} for a large number of sessions. The aim is to extract a few of the most representative principal variation patterns (i.e., principal component functions) from these function curves. For example:

[0088] The first principal component may represent a global pattern where "the risk of failure increases linearly as the conversation progresses."

[0089] The second principal component may represent a bell-shaped pattern where "the risk of failure peaks in the middle of the session".

[0090] The function f_i(t) for each session can be represented as a linear combination of these principal components, and its coefficients (scores) constitute a low-dimensional representation of the dynamic features of the session's failure, which greatly facilitates subsequent clustering and association analysis.

[0091] Dynamic diagnosis and association mining:

[0092] Pattern clustering: Using scores obtained from FPCA, or directly based on functional characteristics (such as peak time, average intensity, and rate of change), sessions are clustered functionally. This can identify session categories with typical failure dynamics, such as "early burst type," "continuously high-risk type," and "late-deterioration type."

[0093] Impact Analysis: The aforementioned dynamic characteristics (such as the intensity f(t) or rate of change f'(t) of a certain type of failure at time t) are used as independent variables, and regression modeling is performed with user behavior indicators in the later stages of the session (such as satisfaction rating, whether to transfer to human intervention, and subsequent complaint intention). This can accurately answer the question, "At which stage of the session, what kind of dynamic failure has the greatest destructive impact on the user experience?", achieving a leap from static pattern recognition to dynamic impact attribution.

[0094] The embodiments of the present invention have the following advantages:

[0095] This invention combines sequence pattern mining algorithms with function data analysis, which can not only identify the type of failure, but also analyze the discrete points of failure types on a timeline, and analyze the impact of these failure techniques.

[0096] 1. In user-AI interaction, failure types are identified through sequence pattern mining algorithms, such as emotional indifference failure, cognitive intelligence failure, and ethical failure, and different failure types can be further subdivided.

[0097] 2. Among the identified failure types, through function data analysis, various discretely identified failure types are placed on the entire interaction timeline to obtain the sequential relationship between different failure types and at what time point the failure is more likely to occur.

[0098] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A dynamic diagnostic method for AI failure modes based on sequence pattern and function data analysis, characterized in that, The method includes: The system acquires raw conversation sequence data generated from multiple rounds of interaction between the user and the artificial intelligence system, and cleans and structures the raw conversation sequence data to form a standardized interactive conversation dataset. Based on sequence pattern mining technology, the standardized interactive session dataset is analyzed, and frequently occurring session patterns that meet preset failure criteria are extracted as candidate failure patterns. The candidate failure patterns are matched and classified with a predefined failure type knowledge base. The failure types include at least emotional indifference failure, cognitive breakdown failure, and ethical anomie failure. The classified failure patterns are then manually or automatically verified. Using functional data analysis methods, an occurrence intensity function is constructed for each verified failure mode on a continuous interaction time axis. The occurrence intensity function is used to characterize the probability or frequency of occurrence of the failure mode at different stages of the session process. The relationship between the occurrence intensity functions of different failure modes on the time axis is analyzed, including temporal sequence relationship, co-occurrence relationship or mutual attrition relationship, and a dynamic diagnostic report on the interaction failure characteristics of the artificial intelligence system is generated based on the relationship.

2. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, characterized in that, Based on sequence pattern mining technology, the standardized interactive session dataset is analyzed to extract frequently occurring session patterns that meet preset failure criteria as candidate failure patterns, including: The sequence pattern mining technology employs the Seq2Pat algorithm, the PrefixSpan algorithm, or a sequence pattern learning model based on a recurrent neural network; the preset failure criteria include a pattern support threshold, a confidence threshold, and a failure keyword matching rule based on semantic similarity.

3. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 2, characterized in that, In the failure type knowledge base, each failure type is further subdivided into at least two subcategories; among them, the subcategories of emotional indifference failure include lack of empathy, inappropriate emotional response, and misuse of emotional vocabulary; the subcategories of cognitive dissonance failure include factual errors, logical contradictions, biased understanding of context, and omission of instruction execution; and the subcategories of ethical failure include biased responses, leakage of privacy information, and generation of harmful content.

4. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, characterized in that, Using functional data analysis methods, an occurrence intensity function on a continuous interaction timeline is constructed for each verified failure mode, including: Normalize the timeline of the entire interactive session; Map each failure mode instance to a corresponding time point on the normalized time axis; By employing nonparametric smoothing techniques or basis function expansion, the distribution of discrete failure mode instances on the time axis is fitted to generate a continuous occurrence intensity curve, i.e., the occurrence intensity function.

5. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, characterized in that, The content of the dynamic diagnostic report includes: A list of core failure modes and their respective types; Intensity trends of various failure modes in the early, middle, and late stages of interaction; A critical failure mode cluster, which consists of multiple failure modes that are highly correlated or frequently co-occur on the time axis; Recommendations for optimizing AI systems based on time-dynamic features.

6. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, characterized in that, The artificial intelligence system includes, but is not limited to, generative large-scale language models, task-oriented conversational robots, emotional companionship AI, or intelligent customer service systems; the original conversation sequence data is in text form, or text converted from voice conversations.

7. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, characterized in that, The automated verification is achieved by training a binary classification verification model, which takes candidate failure patterns and their contextual information as input and outputs a confidence score that the pattern is judged as a true failure.

8. The AI ​​failure mode dynamic diagnosis method based on sequence pattern and function data analysis as described in claim 1, the method further includes: The generated dynamic diagnostic report is correlated with user satisfaction scores, session termination rates, or user complaint data collected concurrently to quantitatively analyze the substantial impact of the dynamic characteristics of specific failure modes on user interaction experience, and the analysis results are fed back to the training or optimization module of the artificial intelligence system.

9. A dynamic diagnostic system for AI failure modes based on sequence pattern and function data analysis, characterized in that, The system includes: The data preprocessing module is used to acquire the raw conversation sequence data generated by the multi-round interaction between the user and the artificial intelligence system, and to clean and structure the raw conversation sequence data to form a standardized interactive conversation dataset. The sequence pattern mining module is used to analyze the standardized interactive session dataset based on sequence pattern mining technology, and extract frequently occurring session patterns that meet preset failure criteria as candidate failure patterns. The pattern classification and verification module is used to match and classify the candidate failure patterns with a predefined failure type knowledge base. The failure types include at least emotional indifference failure, cognitive breakdown failure, and ethical anomie failure. The module also performs manual or automated verification on the classified failure patterns. The function data analysis module is used to construct an occurrence intensity function on a continuous interaction time axis for each type of verified failure mode using function data analysis methods. The occurrence intensity function is used to characterize the probability or frequency of occurrence of the failure mode at different stages of the session process. The dynamic diagnostic report generation module is used to analyze the relationship between the occurrence intensity functions of different failure modes on the time axis, including temporal sequence relationship, co-occurrence relationship or attrition relationship, and generate a dynamic diagnostic report on the interaction failure characteristics of the artificial intelligence system based on the relationship.

10. The AI ​​failure mode dynamic diagnosis system based on sequence pattern and function data analysis as described in claim 9, wherein the dynamic diagnosis report generation module integrates an association rule mining unit for discovering temporal association rules between different failure mode intensity functions, and predicting other failure modes that may be triggered in a specific interaction stage based on the rules.

Images