This paper outlines the basic concept of fault detection and isolation (FDI) in dynamic systems based on analytical process models and briefly reviews the most important approaches. Emphasis is placed upon residual generation using state observation techniques. Because of the strong impact of parameter variations on the applicability and efficiency of model-based FDI, the issues of parameter sensitivity as well as the design of robust FDI schemes are addressed in more detail.
Systems such as chemical plants and nuclear plants have many sensors to detect abnormal events. When some event becomes abnormal, sensors monitoring it make alarms. But too many sensor alarms may prevent the operator from identifying the cause. In design of diagnosis systems, how to choose necessary channels for locating the cause of system failure becomes important. This paper considers the fault distinguishability of diagnosis systems which have many observation items to identify causes of system failure. The diagnosis system must distinguish an abnormal event from the others based on the observation data. A simple method is developed for obtaining minimal combinations of observation items which guarantee the fault distinguishability. Two types of diagnoses are considered: one directly identifies the abnormal event based on all observation data, and the other identifies it step by step from the system level to the component level. Illustrative examples show the detail of the method and its application to the diagnosis of a marine-engine cooling system.
This paper pertains to the problem to locate and estimate a leak occurred in a gas transport pipeline. We have developed an acoustic method which can locate leak sites and can estimate cross-sectional area of the leak from acoustic signals measured at two terminals of the pipeline. We utilize the fact that the occurrence of hole changes the mode of acoustic waves standing in the giant wind instrument (the pipeline), i.e. new standing waves which have the frequencies determined by the leak sites and the amplitude determined by the leak size appearing. The sites and the cross-sectional area of the leak can be calculated from the new acoustic components. Laboratory experiment demonstrated validity of the method.
Since superconducting magnets (SCM) are going to be indispensable to magnetic levitated train, nuclear fusion, magnetic resonating imaging, rotational machines, etc., they must be placed great reliance on its repetitional operations. But without appropriate evaluating methods, these promising techniques must remain still in science levels and hard to be transferable to real human technologies. SCM, being used under dynamical operation with linking other electro-magnetic systems as said above, induce high voltage from which monitoring supercondecting to normal transitional voltage is difficult to distinguish. To solve this problem, monitoring SCM by Acoustic Emission (AE) from themselves, have been found effective, in particular, during the dynamical energizing of them. As for a demonstration, this paper will report mainly how to monitor 3 MJ-SCM and a few results of the experiments aquired both by counting and locational mode of AE in pulsed and repeated operations of the magnet. Some discussions on the AE monitorings are also made along the main issues to be solved in future.
A new sensor for measuring the three-dimensional coordinates and the three-dimensional shape of an object was fabricated. A laser-spotlight was used as the light source. This laser-light is irradiated on a polygon mirror. As the mirror rotates, this light scans the object surface from the lower side to the upper side. The light reflected from the object scans the surface of the sensor developed in this study. On the sensor surface, 21times;21 phototransistor-elements are arranged two-dimensionally. Out of these phototransistor, only the elements receiving light operate and their output signals are sent to the digital parallel processing circuit. The role of this circuit is to detect the position (xs, ys) of the center element among the phototransistors, activated by the light reflected from the object. These detected values are then passed to a personal computer. By processing the values, xs and ys, with the computer program, the three-dimensional coordinates (X, Y, Z) of the center position of the laser-light spot on the object surface are calculated in a real-time fashion. The time required for the measurement of the (X, Y, Z) is as small as 3.3μs. In addition, even when the surface of the object is illuminated by the background light of the magnitude of 1, 000lx, there is no influence on the measured values (X, Y, Z). Furthermore, the data of many positions on the object surface are obtained and stored in the memory units of the computer by repeating the above-mentioned operations, the system can offer a display of the three-dimensional shape of the object on the CRT, but, the number of phototransistor elements are fow 21×21 elements. This causes that the maximum error ΔY (or ΔZ) in the measured value Y (or Z) becomes as large as 3.4mm for the distance of about 320mm between the object and the sensor.
It is known that many cases of failures on ICs were caused by surge. Nevertheless studies on degraded ICs scarcely have been done. When surge energy is not enough to destroy ICs, it is presumed that they would be degraded. We made the degraded analogue ICs in our laboratory. We showed before that using output noise from ICs is useful to detect degraded ICs. In this paper, we cleared the physical mechanism of the degradation. As the result, the surge current paths which energy is not enough to destory ICs, are limited. And the defects arose in PN junction of the transistor which formed the IC. By use of accelerated test with heat stress, spread of degradation area was observed as the time proceeds. We examine the progress of degraded ICs with putting surge again. The degraded ICs was failed by lower surge voltage than the degradation starting surge voltage. Degraded ICs will be failed by discharge again, and hardly failed as the time proceeds.
In this paper, we propose a new system which can reduce to a certain degree the noise component of voice signal. Experimental evaluation of its efficiency is also made. The principle of this system is based on the well known fact that the power spectrum and the autocorrelation form a Fourier transform pair with each other. An original noisy input signal is first processed with FFT, and then the square root of the resultant power spectra is extracted. Next the inverse Fourier transformation is made on it. Only a section of one pitch period of the output is then picked out as a useful part. Successive concatenation of these sections provides the desired output signal. It is easy to find that the final signal contains little noise component. We have applied the present system to the detection of vocal sound elements such as /a/, /i/, /s/ etc., and confirmed successfully its validity through the listening test of normal conversation.
Load forecasting is a basis for power system planning and operation aiming at high reliability and low cost in electrical energy supply. Accurate load forecasts need description of relationship between the load and its determining factors. However there are various factors such as weather conditions and economic situations that may influence the load. It is not easy to discuss general methodology to choose essential factors and represent the relationship exactly. Thus most of the existing methods concentrate their targets on restricted situations or resort to empirical and subjective knowledge. In the paper a method is proposed for one-day-ahead load forecasting, which is applicable to general situations without empirical knowledge. It is a self-organizing method for making a load forecasting model in terms of the determining factors. There is no need for knowing definitely which are necessary factors for forecasting. A simple initial structure of the model is given first, then newly measured data are fed to the modeling process everyday, and a suitable model grows up through the daily forecasting. Even after a suitable structure is derived, the self-organization process continues searching for more suitable one. This is useful and important feature since the load has a time varying character. The method is applied to the daily peak and bottom load. The results show the effectiveness of the method: the self-organized model structures agree in some parts with expected structures from empirical knowledge; and the mean error over one year is around 1.7_??_2.1%.