1 Introduction
2 Methodology
2.1 Axiomatic Design
2.2 Trimming Technique
2.3 Ideal Final Result
2.4 Technology Evolution Theory
3 New Hybrid Axiomatic Design Methodology
4 Case Study
4.1 Defining the Design Problems
4.2 Innovative Process Based on the New Hybrid Methodology
-
Step 1: Define the necessary level of AD syntax.According to the CRRC, the training content used in rehabilitation varies according to the level of dysfunction and physical condition. In this work, we mainly focused on late-stage rehabilitation training and the development of a mobile robot for rehabilitation training. The essential FR (FR1) can be expressed as follows: the design parameters must be appropriate for a lower limb rehabilitation training robot.An appropriate design parameter (DP1) is required to satisfy FR1. Thus, DP1 can be described based on the relevant conversion technology and structure.There is only one design parameter (DP1), and there is no property-related phenomenon. Therefore, we need to decompose FR and DP continuously.According to DP1, the further decomposition of FR1 represents the following FRs:
-
FR11: Provides a proper rehabilitation mechanism for rehabilitation training involving hemiplegic patients;
-
FR12: Able to detect and record gait and fall information;
-
FR13: Able to adjust dimensional parameters and ensure safety;
-
FR14: Provide simple operations with good interactivity; and
-
FR15: Have a low cost.
-
-
The FR hierarchy is constructed from the corresponding DP hierarchy.The alternative solutions and various DPs are analysed based on the new hybrid methodology, as previously noted.
-
Step 2: Calculate the relevance index of each domain, i and j, to determine the design parameters to which attributes are related.The final relevance index for each domain, i and j, is calculated based on the accumulation of weighted subrelevance indices from the abovementioned 3 criteria indices using Eq. (3).where i and j are the indices of the relevance calculation and \({R}_{ij}\) is the final relevance index for DPs i and j, and its value ranges from −1 to 1. \({F}_{ij}\), \({D}_{ij}\), and \({P}_{ij}\) are the three relationship indices for i and j. These indices are related to the functional requirements, design parameters, and properties. \({\omega }_{F}\), \({\omega }_{D}\), and \({\omega }_{P}\) are the weights of the corresponding 3 indices. The weight of each factor is between 0 and 1, and the sum of the 3 weights is equal to 1.$$R_{ij} = \omega_{F} F_{ij} + \omega_{D} D_{ij} + \omega_{P} P_{ij} ,$$(3)
-
Step 3: Iteratively match and merge the design parameters until the DP with the lowest information content is obtained.We can identify the target DPs (according to the following iteration condition: if \({R}_{ij}>0.6\)) that can be integrated or trimmed based on the relevance index and then iteratively combine them. According to FR12, the corresponding DP12 should be able to detect gait and fall information and record the training time, number of steps, and movement speed. We conducted related investigations and research on the existing gait detection technology and summarized several common methods, such as those based on image detection [26], environmental perception-based detection [27], and wearable sensor detection [28, 29].We can use the hybrid model to minimize the total amount of information content. We identified DP12 as the target of iterative combination based on the relevance index. According to trimming technique (1), as discussed in Section 2.2, the IFRs (self-realization) and technology evolution theory (in this case, evolution rule 5: transition to a flexible system or a mobile system to improve controllability), as noted in Sections 2.3 and 2.4, we finally met some detection requirements by arranging the photoelectric sensors at specific spatial locations and applying a novel classifier algorithm. The novel classifier algorithm is detailed in another paper [30‒32]. Compared to other sensor combination methods, the use of photoelectric sensors dramatically reduces the complexity of the design. We can list the design parameters that meet the functional requirements in the design matrix and calculate the information content of the design parameters. The relevance index and total amount of information content between the FR and DP matrices are given in Table 1 based on the new hybrid methodology, as previously discussed.Table 1Information content analysis based on the new hybrid axiomatic design methodologyDP12Alternative solutionsThe final relevance indexInformation content (Recognition accuracy)Image processingWearable sensorPhoto-electric sensorSurrounding environment perceptionTension sensorFR12FR121: Gait detectionImage processingX0.540.350.234Wearable sensorX0.680.136Photoelectric sensorX0.322FR122: Fall detectionImage processingX0.5− 10.420.620.12Wearable sensorX− 10.450.820.089Photoelectric sensorX− 10.820.415Surrounding environment perceptionX0.350.252Tension sensorX0.184FR123: Detecting the training timeImage processingX0.42− 1− 10.252Wearable sensorX0.420.30.029Photoelectric sensorX− 10.007Surrounding environment perceptionX0.515FR124: Detecting the number of stepsImage processingX0.420.56−10.515Wearable sensorX0.780.760.089Photoelectric sensorX0.640.007Surrounding environment perceptionX2.32FR125: Detecting the patient’s walking speedImage processingX0.420.520.320.252Wearable sensorX0.420.30.515Photoelectric sensorX0.250.007Surrounding environment perceptionX2.32In this article, we use the accuracy rate of different alternative solutions as the basis for calculating the information content and set the recognition accuracy rate Ai to calculate the information content using Eq. (4):where TP+FN represents the number of samples that are positive, and P+N is the total number of samples.$$A_{i} = \frac{TP + TN}{{P + N}},$$(4)$$P_{i} = A_{i} ,$$(5)$$I_{i} = \log_{2} \frac{1}{{P_{i} }}.$$(6)As shown in Table 1, we found that for gait detection and fall detection, the solution of wearable sensors has the least information content; for step counting, detecting the training time and detecting the patient’s walking speed, the solution of photoelectric sensors has the least information content.To accurately recognize the various states of the patient in real time, we proposed a multi-sensor system to obtain multiple features and classify activities from different dimensions. The system board was designed with a STM32 microprocessor powered by a 5 V battery. To improve the accuracy of fall detection, we developed a tri-sensor detection system (as shown in Figure 3) for our specific rehabilitation robot. The photoelectric sensors collect the spatial distribution features of the gait for activity recognition. The tension sensors collect the directional features by sampling the difference in the same-side sensor signals. The accelerometer sensor collects kinematic information for activity recognition.×Through the new hybrid methodology, we successfully designed a new low-cost robot that meets almost all the relevant requirements. Compared to the previous rehabilitation training robots, the cost of the new rehabilitation robot is reduced by as many as 42%, and it allows patients to achieve omnidirectional actuation with straightforward manipulation, which significantly improves patient satisfaction.