When there are many variables needed for recognition such as handwritten letters and individual person, the T method having more than one story is introduced here. The total number of variables is classified into g subgroups and calculate the following two statistics in each subgroup as explained in the last chapter

　　　Y₁=slope coefficient β

　　　Y₂=1/√η

The η is standard signal-to-noise ratio. Then we have 2g statistics in each member of unit space as below:

　　　Z₁=Y₁ of the first subgroup

　　　Z₂=Y₂ of the first subgroup

　　　・

　　　・

　　　・

　　　Z_2g=Y₂ of the last subgroup

From these statistics, we can calculate Y₁ and Y₂ as the total slope β and the total standard S/N ratio η in each member of unit space. Using MTA we get final distance of each member of unit space. The distribution of the distance is used for recognition. That is the same as control chart by Shewhart. However x chart and R chart are put to together in multidimensional space.

This study was carried out as a preliminary enabling research for earthquake prediction. When earthquake predictions are made by the MT system from enormous volumes of time-series data etc., the problem of data compression arises. This paper discusses the status of earthquake data, which is an unfamiliar topic in quality engineering, and considers optimal data compression methods for earthquake prediction. The idea of analysis of distribution of periodic data, which is an original concept from on-line quality engineering, was applied to time-series data. Periodic analysis was shown to provide an effective data compression method, enabling the characteristics of earthquake data to be expressed by signal-to-noise ratios with periods of 1/100, 8/100, and 32/100 seconds.

Patients with C6-level cervical spine injuries can move their arms only feebly, if at all, and their grip has no strength. We have developed an upper-arm rehabilitation system that rehabilitates such patients to the extent that they can take meals by themselves in half a year. In this system, a patient traces a target point moving at a constant speed on a monitor screen with a pointer that he can operate by a joystick. The fingers of the patient's hand are placed on the joystick and the back of the hand is secured by a hook-and-loop tape fastener. The joystick is movable in a range of two to three millimeters, with a resilience that stimulates the patient's arm. The patient follows the motion of the target point with his eyes and changes the direction of input to the joystick accordingly. Upper-arm function improves as the training continues. Physical rehabilitation methods are evaluated in different ways for different pathological conditions and training methods. In this study the evaluation procedure was standardized by use of the MTA method to support diagnosis as well as providing rehabilitation benefits for the patient.

The Atractylodes lancea rhizome widely distributed in China is a natural medicine derived from Atractylodes lancea and Atractylodes chinensis in the aster family (Compositae) and is used as an ingredient of many kinds of Kampo medicines. Previously we reported on a componential analysis of the Atractylodes lancea rhizome material in various regions in China carried out by the MTA system, from which we obtained distances that made it possible to estimate characteristic differences, and by which we were able to estimate the longitude of specimens of unknown provenance or origin by an extrapolation formula. In this report, we attempt to solve problems in MTA analysis using north-south differences in latitude as a signal to estimate the latitude of Atractylodes lancea rhizome by using the Taguchi-Schmidt (TS) method and Taguchi (T) method,and estimate regions of distribution in China of Japanese Atractylodes lancea rhizome thought to derive from China. We also report results in which applying to presumption of Japanese Atractylodes lancea rhizome and hybrids between Atractorlodes lancea and Atractorlodes chinensis.

Economic data involve large numbers of variables and show irregular movement over time, making quantification and estimation difficult. Time-serial changes in a large number of variables can be likened to the functioning of the human body. The MT system was therefore applied on a trial basis to corporate financial data, which can be viewed as expressing corporate health. Because financial data take on both positive and negative values, a Schmidt orthogonal expansion was necessary, so the TS method (which forms part of the MT system) was used with operating profit as the target variable. The results demonstrated the feasibility of estimating profits by the TS method, and of indicating the estimation error by the S/N ratio.

In the development of reciprocating piston internal combustion engines, to reduce slap noise caused by collision of the piston against the cylinder wall, slap noise is predicted and remedied by using numerical simulations of piston movements. This simulation was combined with parameter design to optimize the piston shape to reduce slap noise and its variability. Seven design parameters were assigned to an L18 orthogonal array as control factors, and thirteen noise parameters were assigned to an L36 orthogonal array as noise factors. The collision energy of the piston was calculated from a direct product of the two orthogonal arrays, and the signal-to-noise (S/N) ratio and sensitivity were evaluated on a target-is-best characteristic basis. Three parameter designs were carried out with twelve parameters in total by replacing some of the control factors. Confirmation computations under the optimal conditions revealed gains of 1.1 dB in signal-to-noise ratio and 1.8 dB in sensitivity. This effect corresponds to an improvement of 1.8 dB in collision energy and 0.5 to 1 dB in engine noise.