Human factor reliability assessment method for robot-assisted rehabilitation

Abstract

The invention relates to a human factor reliability assessment method for robot-assisted rehabilitation, and the method comprises the following steps: constructing a multi-background expert group, and distributing expert weights; key task selection and key subtask division are carried out on the rehabilitation robot working system; an SHELL model is applied to identify EPCs in the robot-assisted rehabilitation process; matching a corresponding general task type, a related GEP and a corresponding EPC for each key sub-task according to the HEART technology; calculating a score-based direct influence matrix of each expert for the EPC by using an improved Z-pole coordinate system, and aggregating a plurality of direct influence matrixes by considering the weight of the experts to obtain a group direct influence matrix; calculating an evaluation influence proportion for the EPC based on the group direct influence matrix; and calculating the human error probability of each key sub-task based on the evaluation influence proportion. Compared with the prior art, the method has the advantages that the multi-expert weight is considered, and the comprehensive evaluation result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of robot-assisted rehabilitation, and in particular to a human factor reliability evaluation method for robot-assisted rehabilitation. Background Art

[0002] As the scale of systems in various industrial fields increases, the complexity of human-computer interaction also increases. Scientific literature shows that nearly 75% of industrial errors can be directly or indirectly attributed to human factors, while the rest are related to pure system failures. Especially in high-risk fields such as nuclear power plants, aerospace, and transportation industries (including railways, shipping, etc.), human factors have become one of the most critical factors in system safety assessment. Analyzing and controlling the contribution of human factors to system output results can effectively prevent dangerous events. In order to reduce the risks brought by human factors, human reliability analysis (HRA) technology came into being. HRA is defined as a study used to identify, quantify and reduce the probability of human error (HEP) when operators perform specific tasks under specific conditions. HRA research was first carried out in nuclear power plants (NPP) in the 1970s and later expanded to other fields such as marine engineering, aerospace, transportation, and healthcare. With the introduction of more fields, various methods and models of HRA have gradually developed and improved.

[0003] The Human Error Evaluation and Reduction Technique (HEART) is one of the human reliability analysis techniques widely used in multiple industrial fields. It was developed by Williams in 1986 and was originally used to analyze nuclear power plant accidents. Currently, the technique has been applied in multiple industries, including medical environments, transportation, railways, and aerospace. HEART provides a systematic framework that helps to discover and solve the root causes of human errors in complex systems. The significant advantages of this technique are that it is universal, fast, and simple. The HEART technique presupposes that conditions cause people to make mistakes. The technique consists of three main parts, namely, generic task types (GTTs), error-producing conditions (EPCs), and assessment impact proportions (APOAs). The original HEART technique provides 9 GTTs and 38 EPCs. Table 1 provides a qualitative description of the 9 GTTs and provides the corresponding basic error probability (i.e., GEP) for each GTT, ranging from 0.00002 to 0.55, with lower values ​​indicating lower error probability. EPCs are defined as any internal or external factors that may have a negative impact on human performance, such as operator experience, age, and physical fitness. EPCs determine the degree of impact of GTT, and the combination of EPCs and GTT jointly determines the human error probability (HEP) and reliability of the task. Evaluators select EPCs that have a significant adverse effect on task execution, and in order to ensure accurate prediction of HEP, it is recommended to control the number of EPCs associated with each task to less than three. APOA refers to the weight determined based on each expert's judgment of the importance of EPCs, ranging from 0 to 1. However, the shortcoming of HEART is that it does not provide a specific APOA determination method, and therefore faces greater uncertainty.

[0004] Table 1 GTT scale in HEART technique

[0005]

[0006] According to the first Global Burden of Disease Study on Rehabilitation published by The Lancet in 2020, at least one-third of the world's population (2.441 billion people) require rehabilitation treatment during illness or injury. This global scientific survey overturns the common belief that rehabilitation is a medical service that only a few people need. Among them, musculoskeletal diseases are the diseases that contribute the most to the demand for rehabilitation and are one of the biggest reasons why children need rehabilitation services. The imbalance between supply and demand has led to the emergence of rehabilitation robots. Traditional rehabilitation treatment models require high skills from therapists, and at the same time, patients may not be able to afford them due to the large number of treatment sessions required. Robot-assisted rehabilitation technology provides treatment opportunities for most patients at a reasonable cost. At the same time, compared with traditional rehabilitation treatment, this technology can not only reduce the burden on medical staff, but also provide a quantifiable measurement of task performance. Because rehabilitation robots are a young and still developing field, specific safety verification procedures and risk assessment studies are limited. As a robot that interacts closely with the human body, medical accidents in robot-assisted rehabilitation are increasing, and equipment complexity and adverse human factors are the main reasons. As a mature technology that effectively identifies potential risks in the system, HRA can improve the human-machine safety of the entire rehabilitation system by conducting risk assessment on the robot-assisted rehabilitation process. However, research on human reliability assessment in the field of robot-assisted rehabilitation is very limited.

[0007] The EPC scale in the HEART technology fails to fully cover all aspects of the robot-assisted rehabilitation process, so the HEP acquisition method based on the HEART technology is not sufficient to deal with the human reliability research issues in the robot-assisted rehabilitation process. In addition, the research on human reliability analysis (HRA) in the medical system is quite limited, and a systematic human reliability evaluation mechanism in robot-assisted rehabilitation has not yet been formed.

[0008] How to achieve human reliability assessment of robot-assisted rehabilitation based on HEART technology has become a technical problem that needs to be solved. Summary of the invention

[0009] The purpose of the present invention is to provide a human factor reliability assessment method for robot-assisted rehabilitation in order to overcome the defects of the above-mentioned prior art.

[0010] The purpose of the present invention can be achieved by the following technical solutions:

[0011] According to one aspect of the present invention, a method for evaluating human reliability of robot-assisted rehabilitation is provided, the method comprising the following steps:

[0012] Step 1: construct a multi-background expert group and assign expert weights;

[0013] Step 2: Select key tasks and divide key subtasks of the rehabilitation robot working system according to the hierarchical task analysis method;

[0014] Step 3, apply the SHELL model to identify the error-generating conditions EPCs in the robot-assisted rehabilitation process from multiple dimensions;

[0015] Step 4: According to the HEART technique, the corresponding general task type GTT, the relevant basic error probability GEP and the corresponding error generation conditions EPCs are matched for each key subtask; the direct influence matrix of each expert based on the score is calculated for the error generation conditions EPCs using the improved Z-polar coordinate system, and the multiple direct influence matrices are aggregated to obtain the group direct influence matrix considering the expert weights;

[0016] Step 5, based on the group direct impact matrix, calculate the assessment impact proportion APOA for the error-generating condition EPCs;

[0017] Step 6: Calculate the human error probability (HEP) of each key subtask based on the assessment impact ratio (APOA).

[0018] Preferably, the factors considered in constructing the multi-background expert group in step 1 include position, relevant working years and educational background; and the TOPSIS technology is used to determine the weight of each expert.

[0019] Preferably, the process of step 3 includes: expanding the SHELL model into a specific SHELL model of the assisted rehabilitation work system, which takes the central personnel as the core, and identifying the error generation conditions from the dimensions of procedures, rules and training and education in the central personnel-software part; identifying the error generation conditions from the dimensions of places and equipment in the central personnel-hardware part; identifying the error generation conditions from the dimensions of internal environment and external environment in the central personnel-environment part; identifying the error generation conditions from the dimensions of individuals and interactions in the central personnel-other personnel part.

[0020] Preferably, the process of obtaining the group direct influence matrix includes: using the improved Z polar coordinate system, performing the following calculation for the error generation condition EPCs:

[0021] Determine the direct impact matrix of each expert based on the language Z-number;

[0022] Using the improved Z polar coordinate system, the direct impact matrix based on language Z number is transformed into the direct impact matrix based on ZPC;

[0023] Convert the ZPC-based direct impact matrix to a score-based direct impact matrix;

[0024] Based on the ZPCS-PA operator, multiple score-based direct influence matrices from the expert group are aggregated to obtain the group direct influence matrix.

[0025] More preferably, the process of determining the direct influence matrix of each expert based on the language Z number includes:

[0026] The expert group determines a pairwise comparison matrix of a series of EPCs through the language terminology sets S and H, namely the direct impact matrix based on the language Z number, where:

[0027] S={S1,S2,...,S T |t∈N *}={no impact, low impact, ..., high impact, severe impact}

[0028] H={H1,H2,...,H T |h∈N *} = {very uncertain, somewhat uncertain, ..., somewhat certain, very certain}.

[0029] More preferably, the conversion of the direct impact matrix based on the language Z number into the direct impact matrix based on the ZPC is specifically as follows:

[0030] Assume there are two language term sets S = {S1, S2, ..., S T |T∈N *} and H={H1,H2,...,H D |D∈N *}, a language Z number is recorded as Z = (A, B) = (μ A (S t ) / S t ,μ B (H d ) / H d ), where S t and H d All are language assessment information; μ A (S t ) and μ B (H d ) are S t and H d The membership degree, μ A (S t )∈[0,1] and μ B (H d )∈[0,1];

[0031] Convert the language Z number into a coordinate point in the polar coordinate system:

[0032] Z=(L,θ),

[0033] Where L is the polar diameter, θ is the polar angle, μ B and μ A Both represent membership, D and T are the number of terms in the language term sets H and S respectively, t is the ordinal number of the term in the language term set S, and d is the ordinal number of the term in the language term set H.

[0034] More preferably, the conversion of the ZPC-based direct impact matrix into a score-based direct impact matrix is ​​specifically as follows:

[0035] Convert each ZPC into a score, and use the score function to obtain the direct influence matrix of each expert based on the score;

[0036] In the improved Z polar coordinate system, the score function is defined as follows:

[0037] J(L,θ)=k1Lsinθ+k2Lcosθ,J(L,θ)∈[0,1]

[0038] Among them, k1 and k2 are proportional coefficients; L is the polar diameter, and θ is the polar angle;

[0039] For two ideal points M = (1 / severe impact, 1 / very certain) and N = (0 / no impact, 1 / very certain), their score functions are J = (1,π / 2) = 1 and J = (1,0) = 0 respectively; therefore, k1 and k2 are 1 and 0 respectively;

[0040] The above formula is simplified to:

[0041] J(L,θ)=Lsinθ,J(L,θ)∈[0,1]

[0042] Therefore, the score function is expressed as follows:

[0043]

[0044] Among them, μ B and μ A Both represent membership, D and T are the number of terms in the language term sets H and S respectively, t is the ordinal number of the term in the language term set S, and d is the ordinal number of the term in the language term set H.

[0045] More preferably, the process of aggregating the multiple score-based direct impact matrices includes:

[0046] The ZPCS-PA operator is defined as follows:

[0047]

[0048] in, represents the element at row i and column j in the score-based direct influence matrix of the k-th expert; represents the element at the i-th row and j-th column in the score-based direct influence matrix of the t-th expert; express right support and adopt and The similarity between right The details are as follows:

[0049]

[0050] in, is the element located in the i-th row and j-th column in the direct influence matrix based on IZPC of the k-th expert; is the element located in the i-th row and j-th column in the IZPC-based direct influence matrix of the t-th expert;

[0051] definition and The Euclidean distance d between them is as follows:

[0052]

[0053] in, and They represent the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the k-th expert based on IZPC, respectively; and They represent the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the t-th expert based on the improved Z-polar coordinate system;

[0054] and The similarity between them is obtained in the following way:

[0055]

[0056] Finally, we get the group direct influence matrix Z = [z ij ] n×n , specifically:

[0057]

[0058] Preferably, the calculation of the assessment impact ratio APOA for each EPC includes: based on the group direct impact matrix, deriving the total relationship matrix by the DEMATEL method, and calculating ω i The value of each EPC corresponds to ω i That is, the APOA value, where ω i Indicates the importance of the i-th EPC.

[0059] Preferably, the calculation of the human error probability HEP is specifically as follows:

[0060]

[0061] Among them, HEP τ represents the probability of human error in the τth subtask, GEP τ is the basic error probability of the τth subtask, EPC i is the error-generating condition of the i-th critical subtask, APOA i Represents the evaluation impact ratio of the i-th key subtask.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1) The present invention introduces a multi-background expert group into the HEART technology and assigns weights to each expert to avoid the possible perspective limitations of unit field experts; the SHELL model is used to identify the error-generating conditions EPCs in the robot-assisted rehabilitation process from multiple dimensions, and the evaluation information of the error-generating conditions EPCs is quantified using the improved Z-polar coordinate system, and finally the group direct influence matrix is ​​aggregated; APOA is calculated for EPCs, and the human error probability HEP of each key subtask is calculated; the improved Z-polar coordinate system allows the divided evaluation language variables to be free from even number restrictions, and the evaluation rule interval can be freely divided according to the analyzed problem. The comprehensive evaluation result of human reliability achieved after considering the expert weights is more accurate.

[0064] 2) The present invention aggregates multiple direct impact matrices based on the improved Z-polar coordinate system into a group direct impact matrix, which can handle the problem of inconsistent expert opinions and help decision makers obtain a comprehensive evaluation result.

[0065] 3) In view of the problem that the traditional HEART method depends largely on the judgment of a single expert and the experts can only evaluate in the form of direct scoring, which leads to inaccurate evaluation results, the present invention uses the toposis method to determine the expert weights, constructs a multi-background expert group and considers the expert weights, uses the improved Z-polar coordinate system (IZPC) to handle the uncertainty of multiple expert evaluation languages ​​and characterize the expert's confidence level, reduces the expert evaluation information loss and thus improves the accuracy of the evaluation results.

[0066] 4) The present invention expands the SHELL model in the field of robot-assisted rehabilitation, comprehensively identifies the EPCs of the robot-assisted rehabilitation system, and uses the SHELL model as a systematic logical framework for EPCs identification in the HEART method, which promotes the universalization of HEART research and can be flexibly applied to other fields and work systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of the human factor reliability evaluation method in the present invention;

[0068] Figure 2 It is a schematic diagram of the analysis results of the SHELL model of robot-assisted rehabilitation in the present invention;

[0069] Figure 3 Schematic diagram of the improved Z polar coordinate system in the present invention;

[0070] Figure 4 It is a schematic diagram of the HEP distribution curve in the present invention;

[0071] Figure 5 It is a schematic diagram of the sensitivity analysis results of the number of expert groups in the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0073] This embodiment relates to a human factor reliability assessment method for robot-assisted rehabilitation. Figure 1 , taking robot-assisted rehabilitation for cerebral palsy as an example.

[0074] Cerebral palsy is a syndrome caused by non-progressive brain damage and developmental defects from conception to infancy, often accompanied by movement disorders and abnormal posture, which poses a significant challenge to children's quality of life and social participation. Traditional rehabilitation methods such as physical therapy play an important role in improving patients' motor ability and quality of life. However, these methods usually require long-term and high-intensity treatment and are limited by the professional level of therapists and insufficient human resources. Robot-assisted rehabilitation therapy, as an emerging rehabilitation method, has become an important supplement or alternative to traditional methods. Nevertheless, as a machine device that is in close contact with the human body, it is crucial to develop and apply effective risk assessment methods to ensure the safety and effectiveness of the rehabilitation process. The lower limb rehabilitation robot is a robotic gait training system that is used to treat patients with walking disorders caused by diseases of the brain, nerves, spinal cord, muscles or bones. In this example, a volunteer child with cerebral palsy uses Rehabilitation treatment was performed, and we will conduct a human factor reliability assessment of the process to illustrate the proposed method. Previously, informed consent for this study was obtained from the patient and the patient's family.

[0075] The method comprises the following steps:

[0076] Step 1: Establish a multi-background expert group and determine the expert weights.

[0077] Step 1.1, construct a multi-background expert group:

[0078] The human reliability evaluation method of the present invention is based on expert opinions. Rehabilitation robots involve multiple complex fields, such as mechanical design, rehabilitation medicine, control engineering, artificial intelligence, etc. The multi-background expert group can introduce multi-disciplinary and multi-field perspectives to make the problem analysis more comprehensive and in-depth. Different experts have different knowledge backgrounds and experience levels. Such individual differences make each expert's contribution to the final evaluation results different. Experts from different backgrounds will provide different insights and ways of thinking based on their respective professional knowledge and experience, which helps to avoid the perspective limitations that may occur among experts in a single field.

[0079] Step 1.2, assign expert weights: The traditional HEART method depends largely on the judgment of a single expert, and experts can only evaluate in the form of direct scoring, resulting in inaccurate evaluation results. In view of the limitation that the traditional HEART technology cannot handle the expert weight assignment, the approximate ideal solution method (TOPSIS) is used to determine the expert weights. TOPSIS does not strictly rely on data distribution and sample content, and has the advantages of small data loss during calculation, simplicity and convenience, and intuitive geometric meaning. To address this problem, a multi-background expert group is constructed and expert weights are considered. The improved Z-polar coordinate system (IZPC) is used to handle the uncertainty of multiple expert evaluation languages ​​and characterize the confidence level of experts, reduce the loss of expert evaluation information, and thus improve the accuracy of the evaluation results. In order to improve the reliability of the evaluation, assigning expert weights can ensure that experts with higher professional levels and richer experience in specific fields have a greater impact on the final evaluation results than other experts, thereby forming a comprehensive risk assessment result.

[0080] In this study, four experts were selected to form a multi-background expert group. Table 2 shows the information about the positions, relevant working years, educational background and weights of the four experts. The weights of the experts were obtained using the TOPSIS technique.

[0081] Table 2

[0082] Expert No. Job Title Relevant working experience Education Weight <![CDATA[e1]]> Rehabilitation Therapist 8 years Bachelor 0.20 <![CDATA[e2]]> Rehabilitation Doctor 7 years PhD 0.32 <![CDATA[e3]]> Product Engineer 4 years master 0.11 <![CDATA[e4]]> Human Factors Engineering Expert 10 years PhD 0.37

[0083] TOPSIS is a multi-criteria decision-making technique proposed in 1981. This method has a small amount of data loss and a simple calculation process. Its core idea is to sort the alternatives according to their distance from the ideal point and the negative ideal point. TOPSIS technology can solve the shortcomings of traditional HEART technology in expert weight allocation. Its specific implementation steps are as follows:

[0084] Step 1: Create the original evaluation matrix

[0085] Construct the original evaluation matrix of the evaluation individual under multiple indicators. It should be noted that for the indicator scores that need to be obtained through subjective judgment, a 9-level scale is used. Assume that there is an expert group E = {e1, e2, ..., e m} and the index set C = {c1,c2,...,c n}. Score all experts according to each indicator and establish the original evaluation matrix as shown in formula (1).

[0086]

[0087] Among them, x kj represents the score of the k-th expert under the j-th indicator.

[0088] Step 2: Get expert weights

[0089] In order to eliminate the influence of different indicator dimensions, the decision matrix X in formula (1) is standardized by formula (2) to obtain a new matrix R = [r kj ] m×n .

[0090]

[0091] The positive ideal solution and the negative ideal solution are obtained by equation (3).

[0092]

[0093] Among them, K b is the benefit indicator set, K c is the set of cost indicators.

[0094] The Euclidean distance between the score of the i-th expert and the positive ideal solution and the negative ideal solution is obtained by the following formula (4).

[0095]

[0096] The relative closeness between each expert’s scoring result and the ideal solution is calculated by formula (5).

[0097]

[0098] The normalized value (w), i.e., the expert weight, is calculated by formula (6).

[0099]

[0100] Step 2, task analysis: A working system is always composed of several tasks, and the tasks represent the main characteristics of the entire system being evaluated. This example selects key tasks for the rehabilitation robot working system based on the hierarchical task analysis method (HTA), and further subdivides them into several key subtasks according to the hierarchical structure of the working system. For the sake of convenience, key tasks and key subtasks are equated with tasks and subtasks in the following, and these subtasks are the necessary steps to complete robot-assisted rehabilitation. After the task analysis is completed, the corresponding generic task type (GTT) is selected for each subtask based on the qualitative description of the GTT in the traditional HEART method. The robot-assisted rehabilitation process includes 6 tasks, which are further divided into 21 subtasks. The analysis results are shown in Table 3.

[0101] Table 3

[0102]

[0103] Step 3, Identification of Error Producing Conditions (EPCs): EPCs are defined as performance-shaping factors during a task that may negatively affect human reliability during a subtask. EPCs are key parameters for calculating HEP and are combined with GTT. Once the GTT of the subtask being evaluated is determined, the evaluator needs to determine the relevant EPCs for the subtask. In the original HEART, 38 different EPCs were identified. However, some of these originally defined EPCs are only applicable to nuclear environments and cannot accurately describe all scenarios in other fields. Therefore, we combined the characteristics of the robot-assisted rehabilitation system and applied the SHELL model to qualitatively analyze the factors affecting human errors in the robot-assisted rehabilitation process. The results are shown in the figure. Figure 2 A human-centered system was established with the help of the SHELL model to identify EPCs in robot-assisted rehabilitation.

[0104] The SHELL model describes a system consisting of four interacting components, with the central personnel component being the core and most critical part of the entire model. In the model, "S" stands for software, "H" stands for hardware, "E" stands for environment, and "L" stands for other personnel. The SHELL model places the central personnel at the core of the system, emphasizing the importance of human factors in system safety in order to optimize the impact of human factors to the greatest extent. Its advantage is that it can deeply explore the intrinsic connection between humans and other factors, thereby providing a more comprehensive understanding of the root causes and causes of accidents. By analyzing the interactions between the four subsystems of software, hardware, environment, and other personnel and the central personnel system in turn, potential risks and weak links within the system can be identified, and targeted prevention and improvement measures can be formulated. Software refers to procedures, manuals, regulations, and other elements in the system that affect operator behavior and decision-making. These software elements provide guidance and standardization, and have a direct impact on the operator's operations. Hardware includes physical entities such as equipment, facilities, and tools. The design and performance of these hardware components directly affect the operating efficiency and safety of the system. High-quality hardware can reduce operator errors and improve system reliability. Environment generally refers to the working environment and surrounding environment of the operating system, including internal and external environments. The internal environment includes natural conditions such as temperature, humidity and lighting, as well as conditions such as workspace layout, noise level and personnel density. Internal environmental factors affect operator behavior and attention distribution. The external environment refers to political and economic factors that directly affect the introduction, use and regular maintenance of equipment. Other personnel refers to communication and collaboration between individuals. Effective coordination between various personnel is a key factor in the successful completion of tasks.

[0105] Using the SHELL model as a systematic logical framework for EPCs identification in the HEART method promotes the universalization of HEART research, making it applicable not only in the field of robot-assisted rehabilitation, but also flexibly applicable to various other fields and work systems. With this approach, decision makers can comprehensively evaluate system safety and make adjustments based on actual conditions. Based on the SHELL model, the possible causes of accidents in robot-assisted rehabilitation are determined. Putting the central person at the core of the extended SHELL model. In the central person-software section (CL-S), we considered three main modules: procedures, rules, and training and education, which were ultimately divided into five specific projects. In the central person-hardware section (CL-H), it was divided into two main modules, namely, places and equipment, which specifically included seven specific projects. In the central person-environment section (CL-E), we mainly divided the environment into two main modules: internal environment and external environment, which were further divided into nine specific projects. In the central person-other person section (CL-L), individuals and interactions were listed as two main modules, which were subsequently subdivided into five projects.

[0106] Combination The characteristics of rehabilitation robots will Figure 2 The SHELL model in is expanded to The specific SHELL model of the assisted rehabilitation work system is given a maximum nominal value for each EPC according to HEART, as shown in Table 4. The first dimension is the embodiment of the four subsystems in the SHELL model, namely software (CL-S), hardware (CL-H), environment (CL-E) and people (CL-L). In the CL-S subsystem, the second dimension includes procedures, rules and training education, which are then subdivided into five minimum items. Places and equipment are the contents of the subsystem CL-H under the second dimension, which are subdivided into a total of eight minimum items. In the CL-E subsystem, we considered both the internal environment and the external environment, and finally obtained nine minimum items. In the CL-L subsystem, individuals and interactions in the second dimension are finally refined into five minimum items. The above minimum items are the analysis results in the third dimension.

[0107] Table 4

[0108]

[0109] Table 4 (continued)

[0110]

[0111] Step 4, evaluate information processing and aggregation, including:

[0112] Matching and selection of GTT and EPCs: According to the 9 GTTs provided by HEART technology, the corresponding GTT and related GEP were matched for each subtask, and the EPCs that may have a negative impact on its performance were determined for each subtask. These tasks were completed by the expert group through brainstorming meetings. The GTT and related EPCs for each subtask are shown in Table 5.

[0113] Table 5

[0114]

[0115] Evaluation information quantification process: In a complex robot-assisted rehabilitation environment, the data comes from an expert group composed of experts with multidisciplinary backgrounds in rehabilitation-related fields. The members of the expert group have rich experience and professional knowledge, and can independently evaluate EPC under different tasks based on practical experience, work system characteristics, etc. In experience-based methods such as HEART, it is more appropriate to express the subjective judgment of experts with language variables than to directly express numerical values. In addition, the flexibility of language preference is improved, and the loss of expert language information is avoided, thereby improving the accuracy and reliability of the final result.

[0116] Improved Z-polar coordinates (IZPC) are a new form of the Z-number. The Z-number is an ordered fuzzy number pair, usually expressed as Z = (A, B). Z = (A, B) is associated with a real-valued uncertain variable X, where component A is a fuzzy restriction on the value that X can take, and the second component B is a reliability measure of A. This shows that the Z-number can not only handle the ambiguity and uncertainty of natural language, but also consider the confidence level of experts in the opinions given. However, the Z-number lacks intuitive expression of data and is fuzzy in calculation. Therefore, the ZPC originally proposed is used to solve the above defects of the Z-number, but ZPC can only allow the polar coordinate system to be divided into an even number of parts, which means that the number of evaluation language variables of the expert can only be an even number, which brings limitations to the operator in formulating evaluation rules, which may lead to inaccurate final evaluation results. Compared with the originally proposed ZPC, IZPC has a more flexible domain, allowing the number of domains to be divided to be a natural number, and the operator can arbitrarily set the number of language variables according to the system being analyzed.

[0117] In expert systems, experts' evaluations are usually fuzzy and uncertain. Improved ZPC allows experts to express their opinions (such as "important" and "very unimportant") and confidence levels (such as "very certain" and "less certain") through fuzzy language. This fuzzy number pair representation makes the expert's evaluation closer to reality and can better reflect the expert's true views on a certain event. At the same time, IZPC allows the evaluation language variables formulated by the operator to be free from even number restrictions, so that it can freely divide the evaluation rule interval according to the problem being analyzed.

[0118] Definition 1: Assume there are two language term sets:

[0119] S={S1,S2,...,S T |T∈N *} and H={H1,H2,...,H D |D∈N *}.

[0120] Determine the direct impact matrix based on the language Z number:

[0121] The Z number of a language can be expressed as Z = (A, B) = (μ A (S t ) / S t ,μ B (H d ) / H d ), where S t and H d All are language assessment information; μ A (S t ) and μ B (H d ) are S t and Hd The membership degree, μ A (S t )∈[0,1] and μ B (H d )∈[0,1].

[0122] Definition 2: There are two sets of language terms:

[0123] S={S1,S2,...,S T |t∈N *}={no impact, low impact, ..., high impact, severe impact}

[0124] H={H1,H2,...,H T |h∈N *} = {very uncertain, somewhat uncertain, ..., somewhat certain, very certain}.

[0125] like Figure 3 As shown in part (a), the quarter circle OMN is evenly divided into T parts along the curve MN, denoted as S1, S2, ..., S T The divided parts are evenly divided, indicating the membership degree μ A (S t ).

[0126] like Figure 3 As shown in part (b), the quarter circle OMN is evenly divided into D parts along the line segment ON, denoted as H1, H2, ..., H D The divided parts are equally divided, indicating the degree of membership. |OM|=|ON|=1. Therefore, the two ideal points are M=(1 / severe impact, 1 / very certain) and N=(0 / no impact, 1 / very certain).

[0127] Definition 3: There are two sets of language terms:

[0128] S={S1,S2,...,S T |t∈N *}={no impact, low impact, ..., high impact, severe impact}

[0129] H={H1,H2,...,H T |h∈N *} = {very uncertain, somewhat uncertain, ..., somewhat certain, very certain}.

[0130] A language Z number Z = (A, B) = (μ A (S t ) / S t ,μ B (H d ) / Hd ) can be transformed into the coordinate point Z = (L, θ) in the polar coordinate system through equations (7) and (8).

[0131]

[0132] Where L is the polar diameter, θ is the polar angle, μ B and μ A Both represent membership, D and T are the number of terms in the language term sets H and S respectively, t is the ordinal number of the term in the language term set S, and d is the ordinal number of the term in the language term set H.

[0133] Definition 4: In the improved Z polar coordinate system, the score function is defined as follows:

[0134] J(L,θ)=k1Lsinθ+k2Lcosθ, J(L,θ)∈[0, 1] (9)

[0135] Among them, k1 and k2 are proportional coefficients.

[0136] For two ideal points M = (1 / severe impact, 1 / very certain) and N = (0 / no impact, 1 / very certain), their score functions are J = (1, π / 2) = 1 and J = (1, 0) = 0. Therefore, k1 and k2 are 1 and 0 respectively.

[0137] Formula (9) can be simplified as:

[0138] J(L,θ)=Lsinθ,J(L,θ)∈[0,1] (10)

[0139] By substituting equations (7) and (8) into equation (10), the score function can be expressed as follows:

[0140]

[0141] The expert group E = {e1, e2, ..., e m}A pairwise comparison matrix of a series of EPCs, i.e., a direct impact matrix based on the language Z number, is determined by the language term set S = {s1 = no impact, s2 = slight impact, s3 = low impact, s4 = medium impact, s5 = high impact, s6 = severe impact} and H = {h1 = very uncertain, h2 = relatively uncertain, h3 = slightly uncertain, h4 = slightly certain, h5 = relatively certain, h6 = very certain}. T = D = 6.

[0142] In the subsequent expert evaluation information processing part, the evaluation information of expert e1 on subtask B3 is used as an example to illustrate the data processing process. Table 6 shows the direct influence matrix of expert e1 based on the language Z number.

[0143] Table 6

[0144]

[0145] The evaluation information given by each expert in the form of language Z-number can be converted into the corresponding direct influence matrix based on ZPC through equations (7) and (8). The direct influence matrix based on language Z-number of expert e1 in Table 6 is converted into the direct influence matrix based on ZPC, which is shown in Table 7.

[0146] Table 7

[0147]

[0148] Each ZPC is converted into a score by formula (11), and the direct influence matrix of each expert based on the score is obtained. Table 8 shows the direct influence matrix of expert e1 based on the score.

[0149] Table 8

[0150]

[0151] Expert evaluation information aggregation:

[0152] The purpose of this step is to use the ZPCS-PA operator to aggregate multiple score-based direct impact matrices from the expert group to obtain a group direct impact matrix to solve the problem of the lack of a specific method for aggregating expert opinions based on IZPC. The PA operator is an aggregation operator that outputs a single value after inputting multiple values. In order to apply the DEMATEL method in subsequent work, it is first necessary to convert the IZPC-based direct impact matrix into a score-based direct impact matrix according to the score function. Then the proposed ZPCS-PA operator is used to aggregate the score-based direct impact matrices from different experts. Under the premise of a multi-background expert group, this operator can deal with the problem of inconsistent expert opinions and help decision makers obtain a comprehensive evaluation result.

[0153] According to the existing definition of PA operator, the ZPCS-PA operator is defined as follows:

[0154]

[0155] in, represents the element at row i and column j in the score-based direct influence matrix of the k-th expert; represents the element at the i-th row and j-th column in the score-based direct influence matrix of the t-th expert; For right support and adopt and The similarity between right The details are as follows:

[0156]

[0157] in, is the element located in the i-th row and j-th column in the direct influence matrix based on IZPC of the k-th expert; is the element located in the i-th row and j-th column in the IZPC-based direct influence matrix of the t-th expert.

[0158] definition and The Euclidean distance between them is as follows:

[0159]

[0160] in, and They represent the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the k-th expert based on IZPC, respectively; and Represents the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the t-th expert based on IZPC.

[0161] then, and The similarity between them can be obtained in the following way.

[0162]

[0163] Finally, according to formula (16), we get the group direct influence matrix Z = [z ij ] n×n .

[0164]

[0165] It should be noted that Satisfy the following properties:

[0166] (i)

[0167] (ii)

[0168] (iii)

[0169] Considering the expert weights and using formula (16) to obtain the final evaluation result after aggregation, the final group direct influence matrix is ​​shown in Table 9.

[0170] Table 9 Group direct impact matrix

[0171] S22 S31 L11 L21 L23 S22 0.0647 0.0850 0.0654 0.1230 0.0464 S31 0.0445 0.0647 0.0740 0.0714 0.0413 L11 0.0329 0.1932 0.0647 0.0501 0.0204 L21 0.2098 0.1094 0.1299 0.0647 0.0267 L23 0.0856 0.1115 0.0550 0.0262 0.0647

[0172] Step 5, determination of APOA value:

[0173] Based on the group direct influence matrix, the total relationship matrix can be derived and the centrality m can be calculated by the DEMATEL method. i , cause degree n i and importance ω i The value of each EPC corresponds to ω i That is its APOA value. The total relationship matrix is ​​shown in Table 10. The APOA values ​​of all relevant EPCs of subtask B3, namely S22, S31, L11, L21 and L23, are shown in Table 11. With the same method, the APOA values ​​of all subtask-related EPCs can be obtained, as shown in Table 12.

[0174] Table 10

[0175] S22 S31 L11 L21 L23 S22 0.5264 0.6637 0.4961 0.5607 0.2593 S31 0.3725 0.5046 0.4118 0.3768 0.2058 L11 0.3695 0.7763 0.4308 0.3728 0.1858 L21 0.8850 0.8848 0.7261 0.5896 0.2836 L23 0.4535 0.6106 0.3912 0.3200 0.2680

[0176] Table 11

[0177] <![CDATA[m i ]]> <![CDATA[n i ]]> <![CDATA[ω i ]]> S22 5.1131 -0.1008 0.2103 S31 5.3113 -1.5684 0.2278 L11 4.5911 -0.3207 0.1893 L21 5.5889 1.1492 0.2347 L23 3.2458 0.8407 0.1379

[0178] Table 12

[0179]

[0180] Table 12 (continued)

[0181]

[0182] Step 6, obtaining the subtask HEP, including the calculation and sorting of the subtask HEP.

[0183] After the previous steps, the GTT of the task and the impact assessment ratio (APOA) of each EPC are determined, and the HEP value of each subtask is calculated according to formula (17). The results are shown in Table 13.

[0184]

[0185] Among them, HEP τ represents the HEP of the τth subtask.

[0186] Table 13

[0187]

[0188] The final HEP calculation results show that The average HEP value of all subtasks in the assisted rehabilitation process is 3.70E-02. Figure 4As shown in the figure, about 71.43% of the subtasks have HEP values ​​lower than the average, indicating that most subtasks can be completed smoothly. However, there are still 6 subtasks whose HEP values ​​exceed the average, and they are obviously concentrated in the wearing stage, indicating that the reliability of the wearing stage is poor and prone to human errors. In addition, it is worth noting that C2 (letting the patient's feet slowly touch the ground) has the highest HEP. At this stage, the patient needs to maintain balance, and the therapist needs to closely monitor the patient's posture and provide timely assistance. In addition, the transition from the suspended state to the grounded state may cause discomfort to the patient, which may lead to somatic reactions, thereby affecting the effect of rehabilitation training.

[0189] Sensitivity analysis: Sensitivity analysis is used to explore the impact of the number of experts on the final risk ranking of subtasks. The risk ranking of subtasks is based on the comprehensive opinions of experts, and the composition of the expert group will affect the final result. Fewer experts may lead to too single evaluation information, which may reduce the objectivity and reliability of the ranking results; on the contrary, although a larger number of experts can provide richer information and diverse opinions, it may reduce consensus and increase the complexity of information aggregation, which may introduce bias in group decision-making.

[0190] Specifically, six experts were invited to form an expert group, and the number of experts involved in the evaluation was varied to compare the consistency and stability of the final ranking results in different scenarios. Figure 5 The risk ranking of subtasks with different numbers of experts is shown.

[0191] from Figure 5 It can be seen that when the number of experts is less than 4, there are obvious differences in the overall ranking of subtasks, and some subtasks show considerable variability. This shows that when the number of experts is small, individual subjective opinions can have a significant impact on the overall results. When the number of experts increases to 4 to 6, the variability of the subtask ranking is significantly reduced and tends to stabilize. This is attributed to the integration of expert opinions from different backgrounds, resulting in more consistent and reliable ranking results. Therefore, the model proposed in the present invention application effectively and efficiently obtains human reliability analysis results, avoids major deviations due to subjective expert opinions, and minimizes the waste of a large amount of human resources.

[0192] Conclusion: This paper proposes a hybrid HEART approach that aims to enhance the analysis of human factors reliability during robot-assisted rehabilitation. The proposed method innovatively adopts elements from various methods to address the unique challenges posed by this emerging field.

[0193] First, the SHELL model was used to comprehensively identify human errors in the rehabilitation robot system, and then HTA was used to perform task analysis. Recognizing the limitations of individual evaluators in the traditional HEART method, we formed a multi-context expert group and used the TOPSIS method to determine the weights of the experts. Subsequently, in order to collect and process the evaluation information, we combined IZPC with the PA aggregation operator and proposed the ZPCS-PA operator to obtain the group evaluation matrix. This combination not only considers the expert reliability that is not considered by ordinary fuzzy numbers, but also reduces the computational complexity of the traditional Z number. In order to incorporate the interdependence between EPCs into the determination of the final APOA, we adopted the DEMATEL method.

[0194] Applying this method to use A case study of lower limb rehabilitation robot for children with cerebral palsy was conducted to evaluate the risk level of rehabilitation training tasks. The higher HEP associated with specific tasks was verified by actual results. The robustness of the proposed method and its effectiveness in the field of robot-assisted rehabilitation were further confirmed by sensitivity analysis.

[0195] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for evaluating human reliability of robot-assisted rehabilitation, characterized in that: The method comprises the following steps: Step 1: construct a multi-background expert group and assign expert weights; Step 2: Select key tasks and divide key subtasks of the rehabilitation robot working system according to the hierarchical task analysis method; Step 3, apply the SHELL model to identify the error-generating conditions EPCs in the robot-assisted rehabilitation process from multiple dimensions; Step 4: According to the HEART technique, the corresponding general task type GTT, the relevant basic error probability GEP and the corresponding error generation conditions EPCs are matched for each key subtask; the direct influence matrix of each expert based on the score is calculated for the error generation conditions EPCs using the improved Z-polar coordinate system, and the multiple direct influence matrices are aggregated to obtain the group direct influence matrix considering the expert weights; Step 5, based on the group direct impact matrix, calculate the assessment impact proportion APOA for the error-generating condition EPCs; Step 6: Calculate the human error probability (HEP) of each key subtask based on the assessment impact ratio (APOA).

2. The method for evaluating the human factor reliability of robot-assisted rehabilitation according to claim 1, characterized in that: The factors considered in constructing the multi-background expert group in step 1 include position, relevant working experience and educational background; and the TOPSIS technology is used to determine the weight of each expert.

3. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 1, characterized in that: The process of step 3 includes: expanding the SHELL model into a specific SHELL model of the auxiliary rehabilitation work system, which takes the central personnel as the core, and identifying the error generation conditions from the dimensions of procedures, rules and training and education in the central personnel-software part; identifying the error generation conditions from the dimensions of places and equipment in the central personnel-hardware part; identifying the error generation conditions from the dimensions of internal environment and external environment in the central personnel-environment part; identifying the error generation conditions from the dimensions of individuals and interactions in the central personnel-other personnel part.

4. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 1, characterized in that: The process of obtaining the group direct influence matrix includes: using the improved Z polar coordinate system, performing the following calculation for the error generation condition EPCs: Determine the direct impact matrix of each expert based on the language Z-number; Using the improved Z polar coordinate system, the direct impact matrix based on language Z number is transformed into the direct impact matrix based on ZPC; Convert the ZPC-based direct impact matrix to a score-based direct impact matrix; Based on the ZPCS-PA operator, multiple score-based direct influence matrices from the expert group are aggregated to obtain the group direct influence matrix.

5. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 4, characterized in that: The process of determining the direct impact matrix of each expert based on the language Z number includes: The expert group determines a pairwise comparison matrix of a series of EPCs through the language terminology sets S and H, namely the direct impact matrix based on the language Z number, where: S={S1,S2,...,S T |t∈N * }={no impact, low impact, ..., high impact, severe impact} H={H1,H2,...,H T |h∈N * } = {very uncertain, somewhat uncertain, …, somewhat certain, very certain}.

6. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 4, characterized in that: The conversion of the direct impact matrix based on the language Z number into the direct impact matrix based on the ZPC is specifically as follows: Assume there are two language term sets S = {S1, S2, ..., S T |T∈N * } and H={H1,H2,...,H D |D∈N * }, a language Z number is recorded as Z = (A, B) = (μ A (S t ) / S t ,μ B (H d ) / H d ), where S t and H d All are language assessment information; μ A (S t ) and μ B (H d ) are S t and H d The membership degree, μ A (S t )∈[0,1] and μ B (H d )∈[0,1]; Convert the language Z number into a coordinate point in the polar coordinate system: Z=(L,θ), Where L is the polar diameter, θ is the polar angle, μ B and μ A Both represent membership, D and T are the number of terms in the language term sets H and S respectively, t is the ordinal number of the term in the language term set S, and d is the ordinal number of the term in the language term set H.

7. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 4, characterized in that: The conversion of the ZPC-based direct impact matrix into a score-based direct impact matrix is ​​specifically as follows: Convert each ZPC into a score, and use the score function to obtain the direct influence matrix of each expert based on the score; In the improved Z polar coordinate system, the score function is defined as follows: J(L,θ)=k1Lsinθ+k2Lcosθ,J(L,θ)∈[0,1] Among them, k1 and k2 are proportional coefficients; L is the polar diameter, and θ is the polar angle; For two ideal points M = (1 / severe impact, 1 very certain) and N = (0 / no impact, 1 / very certain), their score functions are J = (1,π / 2) = 1 and J = (1,0) = 0 respectively; therefore, k1 and k2 are 1 and 0 respectively; The above formula is simplified to: J(L,θ)=Lsinθ,J(L,θ)∈[0,1] Therefore, the score function is expressed as follows: Among them, μ B and μ A Both represent membership, D and T are the number of terms in the language term sets H and S respectively, t is the ordinal number of the term in the language term set S, and d is the ordinal number of the term in the language term set H.

8. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 4, characterized in that: The process of aggregating the multiple score-based direct impact matrices includes: The ZPCS-PA operator is defined as follows: in, represents the element at row i and column j in the score-based direct influence matrix of the k-th expert; represents the element at the i-th row and j-th column in the score-based direct influence matrix of the t-th expert; express right support and adopt and The similarity between right The details are as follows: in, is the element located in the i-th row and j-th column in the direct influence matrix based on IZPC of the k-th expert; is the element located in the i-th row and j-th column in the IZPC-based direct influence matrix of the t-th expert; definition and The Euclidean distance d between them is as follows: in, and They represent the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the k-th expert based on IZPC, respectively; and They represent the L value and θ value of the element located in the i-th row and j-th column in the direct influence matrix of the t-th expert based on the improved Z-polar coordinate system; and The similarity between them is obtained in the following way: Finally, we get the group direct influence matrix Z = [z ij ] n×n , specifically:

9. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 1, characterized in that: The calculation of the assessment impact ratio APOA for each EPC includes: based on the group direct impact matrix, deriving the total relationship matrix by the DEMATEL method, and calculating ω i The value of each EPC corresponds to ω i That is, the APOA value, where ω i Indicates the importance of the i-th EPC.

10. The method for evaluating human reliability of robot-assisted rehabilitation according to claim 1, characterized in that: The calculation of the human error probability HEP is specifically as follows: Among them, HEP τ represents the probability of human error in the τth subtask, GEP τ is the basic error probability of the τth subtask, EPC i is the error-generating condition of the i-th critical subtask, APOA i Represents the evaluation impact ratio of the i-th key subtask.

Images