© 2018 The Society of Instrument and Control Enginners - SICE. As driver fatigue has become one of the main causes of traffic fatal accidents, more and more scholars are paying more attention to the issue which is related to safety. There are two main types of anti-fatigue driving systems: the first one is to capture driver's face image by camera to estimate the degree of drowsiness based on the change of the image. The second one is to estimate the degree of drowsiness based on the driver's heart rates. This paper develops a system based on heart rate monitoring, which calculates driver's respiration by detecting electrocardiogram, and then infers the driver's drowsiness degree from the respiration changes. Although the system has the disadvantages that is directly contacted with skin and is influenced by the driver's actions, it has lower-cost and higher accuracy than the system that using heart rate directly. After verification experiment, the system is confirmed to have the ability to determine the driver's drowsiness effectively. Hence, the system has contribution to solve the driver fatigue problems.

© 2018 The Society of Instrument and Control Enginners - SICE. Nowadays, the demand for nursing care service has been increasing rapidly with the growth of aging population and presence of chronic diseases. To ease the problem of workforce shortage in healthcare facilities, a novel remote monitoring system is proposed based on wireless communication. The developed system applied collected Received signal strength indicator (RSSI) for acquiring the position of patients and the elderly. Several methods based on RSSI are usually used like maximum likelihood estimation and trilateration. However, accuracies of these methods are not ideal for localization due to abnormal RSSI value caused by random reflection or refraction of radio propagation. In this paper, statistical AP selection method is applied to achieve distribution situation of trilateration results. Then Centroid and Kernel Density Estimation method are utilized to obtain estimation points according to distribution of combination. The experimental result shows that the localization system is sufficient with the improved strategy on positioning.

© 2017 IEEE. With the development of logistics industry recently, the problem of package damage has become more serious. Nonstandard driving can make packages turn into abnormal states. This article presents a package state monitoring system with inertial measurement units (IMUs) which can be easily installed and can detect states of packages correctly. The system uses decision tree method and clustering method to classify states with different features like stable as normal states, while turnover, move, shake, and knock as abnormal states. The experiment results show that the system can distinguish different states of packages with different weights.

© 2017 The Society of Instrument and Control Engineers - SICE. In the traditional indoor positioning systems, because of high cost or various environmental interferences, most existing technology such as GPS, Bluetooth and RFID are not so useful. Pedestrian Dead-Reckoning (PDR) is a kind of low-cost location positioning technique that can get high accuracy by estimating the heading angle of target from acceleration, gyro and compass data. Usually, researchers use compass data to estimate heading angle. However, the compass data is easily affected by external environment, such as electrical equipment or some objects made by iron or steel. On the other hand, the calculation results by the obtained data of acceleration and gyro usually have accumulated error with time. When time comes longer, the bigger error often appears. Some researchers use Wi-Fi based on received signal strength indicator (RSSI) to assist PDR for correction. However, some Wi-Fi Access Points (APs) cannot be used because RSSI can be affected by external factors, such as people moving, wall. In this paper, a novel method that can improve PDR by heading correction is proposed. In this method, heading angle is estimated by calculating the yaw angle with Quaternion from acceleration and gyro data, and optimal APs selection method is used to select suitable RSSI for getting high accuracy.

© 2016 The Society of Instrument and Control Engineers - SICE. Over many years, External Corrosion under Insulation (CUI) scientists and engineers have been recognizing it as a very critical and serious case where inspection and maintenance of pipes are greatly needed. CUI is insidious damage mechanism which causes a large percentage of maintenance expenditures. It is the hidden external corrosion that occurs beneath the jacketed thermal insulation in pipes as a result of water penetration. Most of the petrochemical plants in Japan were built over than 40 years ago, and the problem with this kind of corrosion is localized in areas that are not detected directly by inspection programs. Thus, it is especially hazardous when the insulation covers pipes until it reaches dangerous levels that truly threaten the life of pipes. The purpose of this research is to estimate corrosion rates by using Self-Organizing Map (SOM).

© 2015 Institute of Control, Robotics and Systems - ICROS. When wireless sensor networks are adapted to mobile units, flexible and dynamic network topology should be realized because they are generally used to measure state quantity of fixed objects. ZigBee systems can be used to construct dynamic networks because they support multi-hop communication. A remote monitoring system using ZigBee devices has been developed to support the rescue operations in actual disaster areas or special places outside the scope of mobile phones. The system can trace rescue members' actions and locations by combining some sensors such as IMU, GPS, and barometer, in order to help every member to communicate and enhance the security and work efficiency of rescue teams' performances.

© 2015 The Society of Instrument and Control Engineers-SICE. Wireless sensor networks are generally used to measure state quantity of fixed objects. When they are adapted to mobile units, flexible and dynamic network topology should be realized. ZigBee systems are very useful to construct dynamic networks because they support multi-hop communication. On the other hand, when accidents or large fires occur, rapid and appropriate rescue operations should be taken in the scene. It is very important for such operation that rescue team members constantly communicate with each other in the site. Moreover the base station located near the site should be aware of the members' positions and situations to direct them effectively. A remote monitoring system with ZigBee devices has been developed for the rescue operations in actual disaster areas or special places outside the scope of mobile phones. The system can trace members' actions and locations by using IMU and GPS, and help every member to communicate. It can also enhance the security and work efficiency of rescue teams' performances at the scene.

© 2015 The Society of Instrument and Control Engineers-SICE. Driving assistant systems with cameras and sensors are expected as a solution of safe driving, congestion prevention and driving load reduction and vehicle to vehicle (V2V) communication systems that utilize wireless technology has been proposed as an auxiliary system of the existing systems. On the other hand, as long with the increasing of the aging rate of society, traffic accident of seniors become a serious problem to be solved. Smarty automatic driving system has been developed to realize a smarty platooning control system based on the automatic driving system for senior drivers. In this research, an embedded communication system is developed for small automatic electric vehicle, and an algorithm to avoid transmit collision is implemented in the developed system. Based on the vehicle to vehicle communication, a platooning control system including vehicle interruptions and separations is successfully designed. Some experiment results of several platooning patterns for small EV represent the effectiveness of the developed system.

© 2015 The Society of Instrument and Control Engineers-SICE. In this paper, by using the Android mobile terminal, we make a application of performing autonomous driving control small electric car as the target. Small EV car (COMS) is treated as experimental subjects. This application will help the Elderly, driving poor people, fast, accurate to park the car and call the car coming for users easily. Moreover, it is also considered that it becomes possible to inform the vehicle information for users. For this study, it introduce that how making the communication system with the car. Also using Map management tool 'Mapinfo' for design the path for a small electric vehicle, for example the path to/from parking place. On top of that, we seek a control method to safely autonomous driving on the basis of the data around the vehicle. Also for other cars and pedestrians traveling the parking lot, we aim to correspond to the safety and smooth.

© 2015 The Society of Instrument and Control Engineers-SICE. In the city of Kitakyushu, a quarter more of people are older than 65 years old. Especially a large proportion of elderly people are living on their own. So how to let old people easily access the communal facilities like hospitals need be attended. Therefore, the bad connection between public transportation and home is need to be solved. Based on the survey result and road situation in Kitakyushu city, a new type small single-seat electrical vehicle (sEV) is studied to provide a solution for elder to easily access the public transportation. In this research, low cost and safe automatic driving electrical vehicle based on limited number of sensors is focused.

© 2014 Institute of Control, Robotics and Systems (ICROS). When accidents or large fires occur, rapid and appropriate rescue operations should be taken in the scene. It is very important for such operation that rescue team members constantly communicate with each other in the site. Moreover the base station located near the site should be aware of the team members' positions and situations to direct them effectively. A remote monitoring system with ZigBee devices as a wireless communication method is developed for the rescue operations in disaster areas or special places outside the scope of mobile phones. The system can trace members' actions and situations by using IMU and GPS, and help every member to communicate. It can also enhance the security and work efficiency of rescue teams' performances at the scene.

© 2014 SICE. With the aging driver increasing, the smarty automatic driving system needs to be proposed. We analyze the smarty automatic driving system to propose a smarty platooning control system based on the automatic driving system for elderly person. Firstly, the small automatic electric vehicle and the control devices are explained. Then we introduce the Wireless communication method in our system. Based on the communication between the vehicles, we design related control method and the path recognize method for platooning control system using in urban road. At last we do some experiments to verify the effectiveness of our system.

In Kitakyushu city, more than 25 % people are older than 65 years old. The roads in residential area for the people are very narrow, very steep slope and vulnerable. Some needs assessment for small electric vehicle at some event and community activity was done. Based on the survey result and topographical features, the small electric vehicle for elderly person was selected and was modified and the automatic driving system was built. Automatic driving and platooning using Zigbee or Digimesh to exchange driving data between vehicles were built and tested. © 2014 IFAC.

Rapid and appropriate rescue operations should be taken in the scene when accidents or fires occur. It is very important for such operation that rescue team members constantly communicate with each other in the site, and the base station located near the site should be aware of the team members' positions and situations to direct them effectively. A remote monitoring system with ZigBee devices as a wireless communication method is developed for the rescue operations in disaster areas or special places outside the scope of mobile phones. The system can keep track of members' situations and help every member to communicate, and also can enhance the security and work efficiency of rescue teams' performances at the scene. © 2013 IEEE.

Once disasters, accidents or fires occur, rescue teams turn out and take rescue actions immediately. In such cases, rescue members need to communicate constantly with each other in the site, and it is also important that the members' location and situation are informed to the base station to manage them effectively. In order to enhance security and work efficiency of rescue teams at the scene of disasters, a remote monitoring system which can keep track of team members' situations and help every member to communicate during operation has been developed. In the system, ZigBee devices are used as a wireless communication method, and acceleration data corresponding to members' action is used for detection of their action patterns and locations.

Aiming to realize the safe driving for the elderly people and novice drivers, intelligent vehicle technologies are becoming an important research topic in recent years. While driving, sometimes they cannot handle the car in an emergency due to the poor concentration and reaction. In fact, there are a lot of car accidents caused by such people who occasionally step on the accelerator instead of the brake unexpectedly when intending to have an emergency stopping. This paper describes a recognition method for recognizing driving situation. For supporting driving by traffic situation recognition, real-time detection and robustness are increasingly required. In indoor experiments, the RoboCar was used to verify the recognition methods. The algorithm which determines the routes are able to run safety by the CCD camera was used to recognize the white line on both sides of the road. And the image matching algorithm was used to recognize the traffic signs on the routes. The experimental result with high accuracy was successfully obtained. © 2012 ICROS.

Almost all chemical and petrochemical plants in Japan have been operated for 30 years or more. Leakage accidents in piping often occur in such a kind of plants because of CUI (Corrosion Under Insulation) progress. In the worst case, inner fluid leakage might affect human bodies or the environment. Proper predictive maintenance work based of corrosion rate estimation should be performed to prevent the accidents. In the previous research, corrosion rates have been estimated based on databases constructed by cases of past corrosion. However, the method of previous research has some problems, such as the estimation accuracy is insufficient. In this research, a selection support system which can estimate corrosion rates with high accuracy using information gain ratio and reliability is developed. © 2012 ICROS.

Currently, many plants in Japan have been operated for 30 years or more. Leakage accident in piping may occur because of CUI (Corrosion Under Insulation) progress in such a kind of plants. Proper maintenance should be performed because in the worst case inner fluid leakage might adversely affect human bodies or the environment. In the previous research, corrosion rates have been estimated based on cases of the past corrosion. However, the method of the previous research has some problems, such as the accuracy of estimation is insufficient. In this research, a selection support system which can estimate corrosion rates with high accuracy using reliability is developed. © 2012 SICE.

External corrosion of pipes in chemical and petrochemical plants is developing very slowly therefore corrosion of pipes has been disregarded as maintenance objects. However, many pipes are considerably corroded on the outside surface because many plants in Japan have been operated for 30 years or more. Accidents which relate to corrosions sometimes occurred and led to serious problems and vast losses. Corrosion under insulation (CUI) is extremely difficult to be detected and quantified. The purpose of this research is to estimate CUI potential and corrosion rates by using case database and Group numbering of the continuous condition. © 2011 SICE.

Corrosion is degradation of materials' properties due to interactions with the environments, and corrosion of most metal materials is inevitable. Corrosion damage can lead to fatal environmental consequences like toxic substance leakage, as well as huge maintenance and repairing costs during the suspension of process equipment in chemical and petroleum industries. There are many pipes whose outside surfaces are considerably corroded because many chemical and petrochemical plants in Japan have been used for 30 years or more. Corrosions on the outside surface of pipes progress slowly, and corrosion under insulation (CUI) is extremely difficult to detect and quantify. The proper points of pipe inspection are hard to find. Moreover, the insulation stripping work requires a large amount of time and cost. Besides, the successful managerial methods have not been established. The purpose of this research is to estimate CUI potential and corrosion rates by using case database and Group numbering of the continuous conditions. ©ICROS.

Recently, some accidents which are caused by external corrosion of pipes often occur in chemical and petrochemical plants. It is very difficult to detect points of the pipes where corrosion has considerably developed because the pipes are usually covered with insulations and cannot be inspected directly. Moreover, the length of the pipes is more than 1000km, then it needs huge time and cost to inspect all the deteriorated pipes. Estimation method of inspective pipes based on past case databases has been developed as an inspection support system. The estimation accuracy depends on the number of effective cases recorded in the databases and the soundness of the databases. In this research, three kinds of evaluation indexes of the soundness are proposed.

Corrosions of pipes had been disregarded as a subject of maintenance because corrosions in chemical plants progress slowly. However, many chemical plants were built over 30 years ago in Japan. There are many pipes whose outside surfaces is considerably corroded. Accidents which relate to corrosions may occur and lead to serious problems and vast losses. For this reason, the maintenance is greatly needed. Although, each company takes measures to cope with corrosions, successful managerial methods are not established. The purpose of this research is to know corrosion wastages from outside surface of pipes by estimating a corrosion rate in order to select pipes to be checked. To achieve the purpose, corrosion rates based on corrosion cases are estimated by using the database modeling method. © 2009 SICE.

Operation management and equipment control of plants are carried out based on measurements obtained from flowmeters installed in the plants. Therefore, it is necessary to always observe the flowmeters to detect and diagnose faults, and to correct measurements. However, the fault of the flowmeters, such as bias of the indicated value, can be repaired only during regular maintenance under present circumstances. If the flowmeters with bias of the indicated value can be detected beforehand, reasonable repair plans can be planned. In addition, the bias can be corrected during normal plant operation. A fault diagnosis method using historical database is developed for the flowmeters installed in the plant where the state of the plant always changes by influence of load fluctuation. In the previous study, a design method of the fault diagnosis system that can satisfy specifications required from users about probabilities of a miss-diagnosis and an incorrect-diagnosis is developed. This study proposes a new design and pre-evaluation method based on probabilities of a false-alarm and a miss-alarm, which are more important evaluation criteria. The effectiveness of this method is demonstrated by using Monte Carlo simulations of a tank-pipeline system. © ICROS.

A fault diagnosis algorithm using a signed digraph as a model of a system is useful to real-time diagnosis of failures that occur in a chemical plant. It has been improved so much that it can find multiple origins of failures that occur in the plant at the same time. It is imperative that the diagnosis system be evaluated in the design phase of the system in advance for its practical use. In this paper, an accuracy of diagnostic results using the algorithm for multiple origins has been examined by its application to data obtained by the simulation of tank pipeline systems. The accuracy of diagnosis has been evaluated properly by the size of the greatest set of candidates. © 2006 IEEE.

Build to order (BTO) systems are generally used as systems for corporate to manage production or systems for the consumers to customize components. This paper discusses the BTO system for consumers which can decide the specifications easily. The present BTO systems for personal computers can customize a construction of the personal computers on the web in compliance with users' requests. However, there are many expert consumers who feel that the BTO systems are not satisfy with their requests because components of PC prepared in the system are not many enough to choose. In contrast, beginner users do not have enough knowledge of making selection PC components; therefore it is very difficult for them to use the systems without mistaking. This research develops a BTO system built on the web that includes an information database on a huge number of various PC components and a function that can choose the combination of PC components automatically. As a result, expert users can make a PC of best performance capabilities. Moreover, beginner users become possible to select a construction of the PC easily and automatically only by providing the purpose and the budget. © 2006 ICASE.

A fault diagnosis algorithm using a signed digraph as a model of a system is useful to real-time diagnosis of failures that occur in a chemical plant. It has been improved so much that it can find multiple origins of failures that occur in the plant at the same time. It is imperative that the diagnosis system be evaluated in the design phase of the system in advance for its practical use. In this paper, an accuracy of diagnostic results using the algorithm for multiple origins has been examined by its application to data obtained by the simulation of tank pipeline systems. The accuracy of diagnosis has been evaluated properly by the size of the greatest set of candidates. ©2006 IEEE.

1. In order to select test points for erosion-corrosion of carbon steel pipes, a method was developed to predict the area with maximal shearing stress by use of computational fluid dynamics software.
2. In order to select test points for atmospheric corrosion of insulated pipes, a method was developed to predict corrosion rates by use of information obtained in the design stage, information obtained in the operation stage and information obtained from outlooks of insulations.
3. For the safety of chemical plants, safety design based on the concept of independent protection layers (IPLs) has been introduced in chemical/oil refinery industries. Risk assessment/safety design support system is proposed to develop integrated safety management environment for Chemical plants
4. A method based on support vector machine is proposed for fault detection in time series process data. Support vector machine is one of supervised leaning method. We studied how to use this method in fault detection and assessed its potentiality by using real distillation column time series data.
5. In order to clear and localize the effect of equipment trouble, new methods of fault detection and diagnosis was developed by using weak components in noisy signals. Application of the methods to both simulation and real plant showed that the faults can be accurately detected and diagnosed.
6. A nonlinear modeling technique using historical data during normal operations, which is called DB (database) modeling, has been proposed in order to predict the normal states of a dynamic process. We reported the good results of predictions with the numerical experiments by use of the tank-pipeline system, PVC batch process and the boiler plant simulator.
7. Large parts of oscillations observed in chemical plants are caused by faults of sensors or actuators. In this research, the fault detection capability of the method using high order statistics was examined and analyzed. And a control algorithm to stabilize the control system with abnormal sensors or actuators was proposed.

複数の異常がプラント内に同時に発生している状態下に置いて、「複合異常原因」に対応した符号付有向グラフを用いた異常診断システムは異常原因の候補を多数出力する場合が多く、診断精度が高いとは言い難い。そのため、実用的なレベルでの診断精度の事前評価を行うためには、診断精度を向上させる必要がある。そこで、昨年度は符号付有向グラフを用いた異常診断システムの診断精度を向上させるために、状態変数間の異常の伝搬時間を利用し、複合異常原因に対応したアルゴリズムを用いた診断法を開発した。本年度は、開発した複合異常原因対応アルゴリズムを用いて、異常診断システムの診断精度の事前評価法を開発した。事前評価の手法としては、プラント内に存在する測定器に対して組合せ的に表現可能なパターン(測定変数の値を、異常に高い…「+」、正常…「0」、異常に低い…「-」の定性値で表す)を生成し、プラント内で発生する可能性のある異常状態を全て列挙することにより、その診断結果の全候補集合の中から他の候補集合の部分集合にならない「最大候補集合」を求めることにより診断精度を評価した。最大候補集合の探索では、測定器の数が増えると診断回数が指数関数的に増加し、また、複合異常原因を考慮したアルゴリズムを用いると、単一原因のみの場合と比較してさらに診断回数が増加してしまう。そのため、今回は小規模なタンク8槽からなる配管系(測定器15個)を対象プラントとし、新たに複合異常原因対応の分枝限定操作を開発して複合異常原因対応の最大候補集合の探索を行った。その結果、単一原因と比較すると数十倍の探索回数が必要ではあるが、複合異常原因を考慮した最大候補集合を探索することによって、異常診断システムの診断精度の事前評価を行うことが可能となった。今後は、診断精度の事前評価法を利用した測定器の最適配置問題の解法への応用が期待できる

　本研究では、レスキュー隊員の安全確保と活動効率の向上を目指して、ワイヤレスネットワークを利用したリモート監視システムを基本としたレスキュー活動支援システムの開発を行った。無線システムとして2.4GHz帯ZigBeeを採用し、移動体の位置・高度・運動状態、および隊員の心拍数をモニタリングできるシステムを構築した。その実証実験において、外周総延長距離約1.4kmの広範囲においてほぼ安定（最大パケットロス10％以下）してリアルタイムのモニタリングが可能であった。さらに隊員が持ち歩くポータブルデバイスと各種センサ間をBluetoothで接続するHybrid無線システムを採用し、センサ選択の自由度を高めた。

レスキュー隊救援活動の安全性と作業効率を向上することを目的とし、ZigBeeネットワークを用いて複数センサの融合による位置情報及び活動情報を取得し、移動体を監視するシステムを構築してきた。今年度では、慣性ユニットに加えさらに高低差の補正を行うために気圧センサの追加をした。気圧センサによる高低情報の取得実験を行い、建物内では移動平均によるフィルタで高度の変化を誤差2％以内でリアルタイムに検知することが可能であった。また外乱の影響を受けやすい屋外であっても、大気との接触面積を最適化することでGPSと慣性センサを用いた移動体の位置推定における高低情報の補正に有効であることを証明した。

　現在の石油・化学プラントは老朽化が進んでいるため、プラント内の配管の外面や屋外貯蔵タンクの外面・底板の腐食はかなり進行している場合があり、それらが原因で重大な事故や莫大な損失を生む可能性がある。そこで、配管周りの環境情報を利用し断熱材下で腐食が進行している詳細検査すべき箇所の選定を支援するシステムの開発を行っている。このシステムでは与えられたデータベースも用いて調査箇所の選定を行うため、選定箇所の的中率の精度を高めるためには、データベースの拡充が必要不可欠である。そのため、今年度は配管・タンクの壁面の腐食状態を検査する方法とそのデータの収集法を検討した。　配管やタンクの腐食状態を検査する方法として、比較的検査時間とコストが低くプラント運用中であっても検査ができるAE（アコースティックエミッション）診断法に着目した。AE診断を効率的に行うために、無線デバイスを搭載することにより複数箇所の計測データを長時間連続で取得し、腐食位置を正確に特定する技術と計測データの収集作業を効率化するシステムの開発を検討した。　この無線デバイスの選定および通信性能・適用性を検証するために、共同研究として行っていた小型電気自動車の自律隊列走行のための車車間通信システムの開発を行った。無線デバイスとしてはワイヤレスセンサーネットワーク規格の一つであるZigBeeを使用し、近距離間での通信性能を調べた。　さらに別の適用事例として、消防局のレスキュー隊の救助活動中における隊員の位置把握を目的とした遠隔監視システムのプロトタイプを作成した。実際の作業現場で通信実験を行ったところ、リレー式に複数の無線デバイスを経由することによって約1km離れた箇所のデータをほぼリアルタイムに取得できた。　これらの実証実験の結果から、ZigBee ワイヤレスセンサーネットワークがプラント内のAE診断計測システムへの適用に有効である可能が高いことが分かった。今度はAEセンサーをZigBee無線デバイスに融合し、プラント内での実験を行うことを検討している。

現在、石油化学プラントにおける保温保冷材下の配管外面腐食は重大な問題となっている。配管の外面腐食箇所の推定方法としては、今までに発生した過去の腐食事例をデータベースとして利用し、腐食の発生要因である離散的条件（配管方向、湿潤環境の有無等）と連続的条件（温度、設置高さ等ど）から減肉速度を推定する方法が幾つか提案されている。しかし、従来のニューラルネットワークを用いた手法では、学習データの網羅性に問題があり理論値を超える推定を行うことがあった。また近傍探索より加重平均を求める手法（近傍探索法）では、減肉速度の精度や腐食検査ポイントの的中率が低く、また推定値の範囲が広くなってしまうといった欠点があった。そこで、連続的腐食条件である運転温度、設置高さ、管径を従来の連続値から多段階の離散値へと変換し、同一の区分内でのみ加重平均を取ることにより腐食条件の合致の有無を判断する「区分探索手法」が提案された。さらにこの区分探索によって減肉速度の推算および腐食検査箇所の選定を行う方法が提案された。本研究では区分探索手法を基本手法とし、各プラントから集められた腐食事例データベースを類似の特徴を持つ幾つかのサブデータベースへ分割し、推定箇所の環境因子に最も類似しているサブデータベースのみを使って腐食箇所を推算する手法を新に提案した。腐食事例データベースの分類にはSOM（Self-Organizing Maps）を使用し、従来のデータベースをほぼ同サイズの9種類のサブデータベースに分類した。SOMで分割したサブデータベースを使用することにより、従来法よりも減肉速度の推定値のばらつきを抑えることが可能となり、同等の信頼度で比較した場合推定可能率を15％程度向上させることができ、推算精度も10％程度向上させることができた。これらの結果より、異なるプラント間のデータベースを混用しさらに再分割できることが示され、より広い範囲での腐食条件を元にしたデータベースを有効に使用して減肉速度の推算を行うことが可能となった。

保温保冷材下の配管外面腐食箇所の推定方法として、今までに発生した過去の腐食事例をデータベースとして利用し、腐食の発生要因である離散的条件と連続的条件から減肉速度を推定する方法が幾つか提案されている。しかし、従来のニューラルネットワークを用いた手法では、学習データの網羅性に問題があり理論値を超える推定を行うことがあった。また近傍探索より加重平均を求める手法（近傍探索法）では、減肉速度の精度や腐食検査ポイントの的中率が低く、また推定値の範囲が広くなってしまうといった欠点があった。本研究では、連続的腐食条件である運転温度、設置高さ、管径を従来の連続値から多段階の離散値へと変換し、同一の区分内でのみ加重平均を取ることにより腐食条件の合致の有無を判断する区分探索手法を提案した。さらにこの区分探索によって減肉速度の推算および腐食検査ポイントの選定を行う方法を提案した。本手法は、従来のニューラルネットワークを用いた手法よりも減肉速度の推定値のばらつきを抑えることが可能となり、原理上理論値を超えるような出力はあり得ない。また従来の近傍探索法と比較しても、減肉速度の推定精度は大きく変わらないものの、推定可能率を20％程度向上させることができ、従来は推算不可能であった場合においても推算が可能となった。これらの結果より、異なるプラント間のデータベースを混用できる可能性が向上したと考えられ、より広い範囲での腐食条件を元にした減肉速度の推算が可能となった。

　石油・化学プラントでの運転管理や設備管理は、プラント内の流量計から得られる測定値を基に行われる。長時間の運転中には流量計の指示値が正常値から外れてしまうことがあるため、流量計の故障を検出して測定値の修正を行うことが望まれている。しかし、流量計は運転中に取り外して検査・修理を行うことは不可能であり、一般には定期修理時にしか検査・修理できない。　運転中に測定値がずれている流量計を見つけ出すことができれば、定期修理中にそのような流量計だけを検査してその他の無駄な検査をなくすことが可能であり、また測定値のずれの大きさを推定することができれば、定期修理を行うまでの間、測定値を補正して運転を行うことができる。　　本研究では、故障した流量計が複数存在するときの故障診断法の開発及び多重故障診断システムの設計を目指し、以下の研究成果を得た。1.多重故障診断法の開発　検出可能な故障流量計の数を最大4個と仮定し、故障流量計が複数存在している状況下でも高精度に故障流量計を検出するシステムを開発した。2.多重故障診断システム設計問題の定式化　診断の精度の評価基準として誤報確率および欠報確率を用い、ユーザの要求仕様（検出する故障誤差の大きさ等）を満足することができる多重故障診断システムの設計法を確立した。

　石油・化学プラントの運転管理、設備管理はプラントに設置されている流量計から得られる測定値を基に行われるので、常に流量計の故障を検出・診断をして測定値の修正を行うことが必要である。しかし、流量計は運転中に取り外して検査・修理を行うことは不可能であり、定期修理時にしか検査・修理できない。そこで、運転中に測定値が偏っている流量計を見つけ出すことができれば、定期修理中にそのような流量計だけを検査してその他の無駄な検査をなくすことが可能であり、また測定値の偏りを測定することができれば、定期修理を行うまでの間、測定値を補正して運転を行うことができる。　今までの研究で、運転状態が常に変動しているプラントでの運転時における正常状態に時系列データを蓄積した“履歴データベース”を利用して、故障した流量計が１つだけの場合と仮定しての「故障診断法」が提案されている。よって本研究では、故障した流量計が多重存在するときの故障診断法の開発及び多重故障診断システムの設計を目指し、以下の研究成果を得た。1.多重故障診断法の開発　検出可能な故障流量計の数を3個以内と仮定し、多重故障(3個以下)が存在しても高精度で故障流量計を発見するシステムの開発。2.多重故障診断システム設計問題の定式化　上記の多重故障診断法を使用し、ユーザの要求仕様を満足することができるような多重故障診断システムの設計法の確立。　この多重故障診断システムの設計法を利用することにより、診断の対象となる流量計の特性やユーザの要望（欠報率及び誤報率に着目する）から最適な設計パラメータ(最適な誤差二乗和のしきい値と履歴データベースと比較する観測データの長さ)を設定できるようになる。

　本研究では、負荷が変動する診断対象プラントにおける異常の検出手法であるデータベースモデリング法の検出速度の改良を行った。データベースモデリング法の利点は予測精度が非常に高いという点であるが、規模の大きな診断プラントにおいて異常の検出・診断を行う場合には予測結果の探索にかなりの時間を要するためにオンラインでの予測が困難となる問題点があった。　大規模な対象においてデータベースモデリング法を用いるためには、データベースレコード探索を高速することが必要であるが、本研究ではソフトウェアによる効率化として検索アルゴリズムの改良を検討した。　従来のデータベースの検索に時間を要していた理由は、膨大な量のデータ全てに対して入力レコードと類似性の高いレコードを検索していたためである。ここで、対象としている化学プラントの状態変数は比較的揺るやかにしか変化しないという特徴を利用することによって、検索レコードの範囲を予め絞り込むことを考えられる。また、オンライン予測を行う際にレコードデータの入力の待ち時間を利用して、次に入力されるレコードの範囲を予め予測しておき、その予測レコードに対応した検索レコードの絞り込みを行えば、さらにレコード検索を高速化することが可能である。　これらの検索高速化アルゴリズムを配管系プラントの1年間分の測定データに適用し、その有用性を検証した。

　本研究では、負荷変動を伴う化学プラントにおける異常検出の手法であるデータベースモデリング法の改良を行った。従来のデータベースモデリング法は予測精度が非常に高いという利点を備えているが、規模の大きなプラントを異常の検出・診断の対象とした場合には予測結果の探索に時間を要するためにオンラインでの予測が難しいといった問題点があった。　大規模系を対象としたデータベースモデリング法を実用化するためにはまずデータベースレコード探索を高速することが不可欠である。そのためには、ハードウェアの強化とソフトウェアの効率化の両面からのアプローチが考えられるが、本研究では比較的実装が容易な前者を採用した。　データベースの検索は膨大な数の整数演算の繰り返しであるので、ハードウェアの処理能力を向上させれば探索速度はリニアに向上することが期待できる。そこで、データベースを複数に分割し、ネットワークで接続された高速なＨＤと大容量のメモリを搭載した複数のパソコン上でそれぞれのデータベースの検索い、個々の探索結果を基に全体の予測値を求める手法を開発した。この手法によって接続したパソコンの台数分だけ探索速度を高速化することが可能である。実際に４台のパソコンで分割したデータベースをオンメモリで実装し、ネットワークを介してサーバにデータを集約するシステムで実験を行ったところ、パソコン単体の探索速度の3～3.5倍程度まで速度向上が可能であった。　本手法を用いることにより、対象の規模に合わせて探索に使用するパソコンの台数を増やせば、容易にオンライン異常検知が可能なデータベースモデリング法を構築することが可能である。　今後は、探索アルゴリズムの改良化によって探索に使用するパソコンの台数を削減する方法を検討する。