Improved Monitoring and Surveillance through Integration of Artificial Intelligence and Information Management Systems

The paper describes the results of a project which aims to improve the capabilities of an information system (IS) which supports the management of safety of dams. The improvement has been achieved through the incorporation of additional components developed using artificial intelligence concepts and technologies. We describe the pre-existing IS (comprised of automatic monitoring systems, telemetry and data bases), identify user requirements driving the evolution of the IS and explain how AI concepts and technologies may contribute. We describe the functions, the architecture and the AI techniques of two systems (MISTRAL and DAMSAFE) added to the IS. Moreover we discuss the issue of integration of the AI components and the pre-existing system and we present the technology developed to support this process. Finally we give the implementation status of the project (which delivered components operational since 1992) and some information about the user acceptance, development effort and applicability to other fields. 0. The context This paper derives from a project which aims to improve the capabilities of an existing information system supporting the management of safety of dams. The improvement has been achieved through the incorporation of additional components developed using AI concepts and technologies. Significant resources are expended in the field of civil engineering in managing the safety of large structures like dams. In Italy, the collection, storage and analysis of information concerning a dam is considered as a critical part of managing safety. Automatic instrumentation and data acquisition systems are used to monitor the real time behaviour of dams. The output of monitoring systems is presented locally to dam wardens to alert them to possibly dangerous situations. Telemetry systems are used to send the information to a central data base where experts evaluate the status of the structure through the interpretation of data. The difficulty associated with this data interpretation (both at local and at central level) is due to different factors. Among them: the large amount of data; the uncertainty and incompleteness of information; the need for engineering judgment, knowledge of the particular structure, experience of the behaviour of dams in general and a background of general engineering knowledge in order to interpret the data. Marco Lazzari, Paolo Salvaneschi, "Improved Monitoring and Surveillance through Integration of Artificial Intelligence and Information Management Systems", Proceedings of the Tenth IEEE Conference on Artificial Intelligence for Applications (CAIA '94), San Antonio, Texas, 1994 © 1994 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. AI concepts and technologies can assist engineers in safety management [1, 2] by providing additional components to the existing information system such as real time interpretation systems linked to the data acquisition units and intelligent data bases supporting the off-line management of information and interpretation. The paper reports the results of the work done in this field during the last four years at the software technologies unit of ISMES. The paper is organised as below: Chapter 1. describes the pre-existing information system comprised of automatic monitoring systems, telemetry and data bases with associated tools. Chapter 2. identifies the user requirements driving the evolution of the information system and explains how AI concepts and technologies may contribute. Chapter 3 illustrates in some detail the functions, the architecture and the AI techniques of the two systems added to the preexisting information system. Chapter 4. discusses the issue of integration of the AI components and the pre-existing system and presents the technology we developed to support the integration process. Chapter 5. describes the implementation status of the project which delivered components operational from mid-1992, the user acceptance and some detail about the development effort. Finally the scope of the project and the applicability to other fields are discussed. 1. The pre-existing information system A basic requirement of managing dam safety is the monitoring of the structure in order to collect data which are then interpreted so as to understand the state of the dam. In Italy, this is done through a two level organisational structure: a first level identifies and manages possible alarm conditions and, if necessary, calls the second level; the second level manages the available information concerning the dam and evaluates the safety of the structure on a periodical basis or when requested by the first level. To provide for the requirements stated above, the pre-existing information system developed by ISMES was comprised of the following subsystems: a real time automatic monitoring system, a telemetry system and a central data base with associated processing and representation functions. The monitoring system (called INDACO) collects data such as displacements of the dam, basin level, seepage losses, uplift pressures and tests them against thresholds or values predicted by theoretical models (on-line checks). The resulting warnings and associated data are presented to the people working at the first level of the organisation (e.g. the warden of the dam). The data collected by the monitoring system are sent through a telemetry system to a central computer and are loaded, together with manually collected data, into a data base (called MIDAS), which allows the subsequent analysis and interpretation of the behaviour of the structure (off-line check). MIDAS has been operational since 1985 and manages the data of 200 dams in different countries (MIDAS has been sold to different organisations in Italy and abroad). MIDAS is running in UNIX and VMS environments. INDACO, a PC-based system, has been installed at 35 sites for dam monitoring. 2. User requirements and AI contribution The limits of the above described technology and the user requirements for improvement may be classified as belonging to two different levels: local level (management of warnings) and central level (periodical safety evaluation or deep analysis on demand). At local level, monitoring systems currently available allow the carrying out of two kinds of checks on the values gathered by each instrument:  comparison of the measured quantity and of its rate of change with pre-set threshold values;  comparison of the measured quantity Marco Lazzari, Paolo Salvaneschi, "Improved Monitoring and Surveillance through Integration of Artificial Intelligence and Information Management Systems", Proceedings of the Tenth IEEE Conference on Artificial Intelligence for Applications (CAIA '94), San Antonio, Texas, 1994 © 1994 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. with the corresponding value computed by a reference model. Therefore, these checks neither deal with more than one instrument at a time, nor with more than one reading at a time for each instrument. In addition, any behaviour (either of the structure, or of the instruments) which is not consistent with the reference model generates a warning message. Because of the limited interpretation skills of the people on-site, false alarms cannot be identified and therefore require expert attention. At this level, AI may contribute in collecting the expert knowledge related to data interpretation and delivering it through a system linked with the existing monitoring system. The system can filter and classify the anomalies by using different types of knowledge (e.g. geometrical and physical relationships). It can take into account the whole set of measurements and warnings to identify the state of the dam and to explain it. This allows a part of the expert interpretation to be performed on-line, and therefore to reduce the requests for expert intervention and to increase the level of the safety of the dam. At the central level, the existing system contains quantitative data coming from the monitoring systems. The safety evaluation requires the availability of additional types of information (qualitative data coming from visual inspections, results of experimental tests, design drawings, old interpretations) and different types of knowledge (empirical relationships, numerical models, qualitative models), related to different areas of expertise (structural behaviour, hydrology, geology). As a consequence, there is a need for a cooperative supporting system, able to help people in managing the complexity of the evaluation. AI can be helpful through providing new ways to model the behaviours of the physical systems (e.g. a qualitative causal net of possible physical processes). This modelling approach is a useful way to integrate different types of knowledge (qualitative and quantitative data and models) providing a global scenario to understand the physical system behaviour. Moreover AI may be helpful in collecting and formalising different types of knowledge from different experts for data interpretation and evaluation of the dam state. 3. MISTRAL and DAMSAFE To improve the capabilities of the information system in accordance with the above stated requirements, we developed two systems using AI concepts and technologies. The first system, called MISTRAL, operates in real time and is linked to the existing monitoring system. It provides a global interpretation and explanation of the dam state and evaluates it against a desired state. The second system, called DAMSAFE, may be interpreted as the evolution of the MIDAS data base. It integrates the MIDAS system (data base, graphical and computational tools) and adds new types of information (results of visual inspections and tests, design drawings) It provides additional models of the dam system (a causal net of possible processes occurring in the dam and near environment) and tools to support the interpretation of data and the evaluation of possible scenarios using the causal net. Below we describe in some detail the functions, the structure and the AI concepts and technologies related to MISTRAL and DAMSAFE. More details may be found in [3, 4, 5].