Ontology to Support Knowledge Representation and Risk Analysis for the Development of Early Warning System in Solid Waste Management Operations

A knowledge acquisition process was applied for the development of knowledge based early warning system in material recovery facilities. During this process an ontology that describes a typical material recovery facility was developed. The development of the ontology facilitated the knowledge acquisition process of incidents and accidents. This paper, will demonstrate a way to represent the knowledge about incident and accidents that is stored within Fault Tree Analysis diagrams to ontologies. This kind of representation is of importance because it can make the acquired knowledge reusable and easily sharable among computer agents and stakeholders. 1 INDRODUCTION During the operation phase of many engineering systems, a considerable number of problems, faults, incidents and accidents can occur leading to direct and indirect consequences ranging from citizen complaints and increased operational cost to human lives losses and possibly to disasters. In order to retain an operation mode that is considered “normal” the engineers are using models and techniques from a wide range of principals like risk and barrier analysis, cognitive analysis, psychology, ergonomics, computer-human interaction, etc. They are aiming to design better and safer facilities and proper operating procedures to minimize the number of accidents and harmful-contact incidents. Despite all these efforts, many incidents and accidents are still happening. In many, if not in all, of these cases the timely warning and response of imminent problems is more desirable in terms of economic, political, environmental, and human resources than to deal with the outbreak and aftermath in an ad-hoc manner. International Symposium on Environmental Software Systems (ISESS 2007), Crowne Plaza Hotel, Prague, Czech Republic May 22 25, 2007 (Proceedings in CD). In any engineering facility both managers and personnel have to receive and understand the signals that are transmitted by the components of the system and by the surrounding environment indicating potential occurrence of unwanted events. Based on these signals the personnel must react accordingly in order to prevent the unwanted events from occurring. In this framework, computer systems can help managers and personnel (especially the inexperienced) to prevent operational problems, accidents and failures by informing them about the potential unwanted events in a timely manner, by delivering a clear message to stakeholders, and by providing a list of emergency response procedures. Computer based early warning systems could be of use in engineering facilities and in particular in landfills, materials recovery facilities, and incinerators. These facilities are key components of solid waste management systems that are currently in operation in many countries. In addition, these are complex facilities and must oblige a range of environmental, economical, organizational, health/safety, and sociopolitical specifications. Solid waste management systems are involved in a large number of accidents due to poor operational practices in each solid waste management facility. Some of them can be classified in the category of disasters like the one in the Leuwigajah dumpsite in Indonesia [Fricke et al., 2005], where after 3 days of heavy rainfall 2.7 million m waste started sliding down the valley. The waste covered an area of 900 x 300 meters, 147 people died in the ruins of two settlements, and the surrounding environment has been damaged significantly. A more recent example is the fire that burst out in the second larger landfill in Greece the summer of 2006. Most probably the fire was burning in the compacted volume of waste under the subsurface of the landfill for days. It was expanded at the surface after the collapse of a large pile of waste. The fire was burning for 10 days and released large amount of dioxins in the atmosphere. A number of people were seeking medical attention for breathing problems. The incident resulted in a local scale environmental disaster. In short, the consequences of the operational problems of solid waste management facilities, depending on their nature and severity, range from minor infrastructure damages or simple nuisance problems to critical events, which can lead to the loss of human lives or even to disasters. The research goal is to develop an early warning system in engineering facilities that will be able to estimate the possibility of occurrence and/or the probability of operational problems during operations and to provide advice on how to prevent them. A high priority goal is to define the operational problems, their causes, and also the mechanisms that connect causes with operational problems. In essence, a very important sub-goal is to define the complicated picture of the coincidences that can trigger operational problems in engineering systems in a manner that can be sharable, reusable and easily updatable. As a case study, a facility that sort and process household and commercial waste commonly known as material recovery facility has been selected. A material recovery facility is defined as: A central operation where source segregated, dry, recyclable materials are sorted, mechanically or manually to market specifications for processing into secondary materials [Gladding, 2002]. Main reason to select this type of facility are the statistics which have shown that the overall accident rate for the waste industry in the U.K. during 2001-2002 was estimated to be around 2,500 per 100,000 workers [HSE, 2004]. This rate was about four times that year’s national average. In particular, for scrap and material recovery facilities the rates of incidents and accidents are not encouraging. In the 2004-2005 U.K. statistics of fatal injuries [HSC, 2005], the industry with the highest rate of fatal injury to employees was the recycling of waste and scrap, where the rate was approximately 27 times the national average. These statistics are revealing the large size of occupational health and safety problem in the recycling industry, and point out the need for better and safer practices during the operational phase. Early Warning Systems A universal accepted definition of an early warning system does not yet exist − probably one never will [Glantz, 2004]. The amount of truth of this statement is significant. A literature search for early warning systems identifies thousands of hits. Almost all of the references had to do with financial systems for third world countries, tracking the destructive nature of violent conflicts that led to human suffering, or systems for syndromic surveillance and also with human health and traffic systems for the prevention of accidents. This indicates that different perspectives of the term do exist among the scientific community. The United Nations defines EWS as the provision of timely and effective information, through identifying institutions, that allow individuals exposed to a hazard to take action to avoid or reduce their risk and prepare for effective response [ISDR-UN, 2003]. The objectives of such systems should be to provide timely warning of imminent dangers so the managers and personnel can have time to prepare and act accordingly to avoid it. The alternative is to take mitigation actions, and thus to reduce the possibility of loss of life, personal injury, damage to property and loss of efficacy. According to the literature, the four following items are the key elements of a complete and effective EWS [EWC III, 2006]: 1) Risk Knowledge, 2) Monitoring and Warning Service, 3) Dissemination and Communication, 4) Response Capability. These elements are important to early warning systems when it comes to coping with hazardous natural phenomena like earthquakes, tsunamis, floods and droughts. Nevertheless, these elements can be used as a guide for the design and development of early warning systems dealing with problems and accidents during the operations of engineering systems. This paper is focused on the first key element of the list shown above. Unquestionably, the identification of the risks and problems faced by systems, and the understanding of the mechanisms that connects causes with problems is a very important element of any early warning system. In order to identify and analyse these risks in the framework of engineering systems expertise utilization is required. However, expertise on managing and operating solid waste management facilities is typically scattered. In addition, the knowledge on their operations varies among countries and among types of facilities. Unfortunately, few expert operators and managers are recording their experiences and a small number of researchers are investigating systematically the mechanisms and the causes behind operational problems. Thus, in order to identify and analyze the operational problems of material recovery facilities a knowledge acquisition process had to be applied to develop a knowledge base for the early warning system. This paper is focused in that knowledge acquisition process. Particularly, it will illustrate a way of transforming the knowledge about incidents and accidents that is stored in a widely used diagrammatic technique in ontologies. This mapping is of importance because it can make the acquired knowledge easily reusable and sharable among stakeholders, agents, and among other early warning systems developers. 2 CONCEPTUAL DESCRIPTION OF THE EARLY WARNING SYSTEM Before we focus in to the knowledge acquisition process, it is important to make a conceptual description of the early warning system. The early warning system will be consisted of the following software components: 1) Expert system, 2) Web site and 3) Database. The expert system component will be consisted of subcomponents, such as the knowledge base and the inference engine. These components, when combined, will be able to mimic the reasoning of domain experts in material recovery facility operations. The knowledge base will contain the domain experts’ knowledge that has been acquired through the knowledge acquisition process (this process will be presented in more detail the next paragraph). The inference engine will receive real time data from a set of sensors but also will be able to request some extra information from the manager. Based on the received information and based on the knowledge that will be stored in the knowledge base the inference engine will reason and ultimately will provide the early warning of an upcoming unwanted event, together with a set of response actions that will act as a barrier to its occurrence. Useful data will be stored in database tables (proper operation practices and procedures to in case of specific emergencies). A subcomponent of the wed site can provide the user interface module of the early warning system. Other subcomponent of the wed site can provide other kind of services such as explanation on how to use each feature of the early warning system (something analogous to user documentation in software engineering). Information relevant to the development process of the early warning system. On line forums, wikis, and mailing lists that can facilitate the transfer of tacit knowledge during early warnings and emergencies. 3 DESCRIPTION OF METHODS AND PREVIOUS WORK A set of methods have been used during the knowledge acquisition process. These are briefly described bellow. Ontology An ontology defines a common vocabulary for researchers who need to share information in a domain [Noy et al., 2001]. A widely used definition states that an ontology is a formal specification of a shared conceptualization [Gruber, 1993]. It is consisted of definitions of concepts, relations and rules about a domain. Ontologies are widely used in knowledge engineering and artificial intelligent, in different applications of computer science and in new emerging fields like the semantic web. An ontology can be used in knowledge based systems with the potential to employ inference and can be build based on artificial intelligent modeling techniques like frames and first-order logic, as well as based on description logic modeling techniques. Software engineering techniques like UML and databases techniques like Entity Relationship diagrams can also be used to build ontologies. All these knowledge modeling techniques can not represent the same knowledge with the same degree of formality and granularity. However, it is important to remark that the model can only be considered an ontology if it is a shared and consensual knowledge model agreed by the community [GómezPérez et al., 2003]. FMEA Failure Mode and Effects Analysis (FMEA) is a qualitative risk and reliability analysis method. It is usually applied during the early phases of a product development life cycle. It allows a systematic analysis of a variety of failures and also allows assessing their unwanted effects. In order to perform a FMEA the following steps have to be made. 1) Identify the component or the functions of the product, 2) Identify potential failure mode for each component or function, 3) Identify potential failure effects for each failure mode, 4) Determine the severity of all effects, 5) For each failure mode identify potential causes, 6) Determine the frequency of each failure mode. The information that has been gathered from this process is stored in a table format forming FMEA tables. FTA Fault Tree Analysis (FTA) is a widely used probabilistic risk and reliability analysis method. It represents graphically the relations of the undesirable events of a system, which are described by the term “top events”, with their causes, which are described by the term “basic events”, via logic operators or gates (AND gate – OR gate). In order to perform a FTA the following steps have to be made. 1) Select a top event for analysis, 2) Describe all events which immediately cause the top event, 3) Define the logic gate that connects the top event with the immediate events, 4) For each event defined in the previous step continue describing its immediate causes and corresponding logic gate until the granularity level of the analysis is reached. The information gathered from this process is stored in tree like diagram known as FTA diagram (see Figure 1 below). Previous Work It has been mentioned earlier, that ontologies have been used to represent and edit the domain knowledge in different applications. However, a limited number of papers have implemented ontologies together with risk analysis techniques. One attempt made by [Lee 2001] presented an approach to build diagnostic models bringing together FMEA and ontologies. Another attempt that made by [Dittmand et al., 2004] introduced a top-down approach to define concepts in FMEA tables using the F-logic ontology language. Finally, an approach to produce FMEA tables from an ontology, was addressed by [Koji et al., 2005]. In this case a knowledge transformation system was developed. It was composed by an extended functional ontology used to define concepts in extended functional models, a FMEA ontology used to define concepts in FMEA tables, a mapping knowledge ontology used to specify the correspondence between similar concepts in the previous ontologies, and an transformation engine that used XSLT style sheets to produce the FMEA tables. These techniques have demonstrated that knowledge derived from FMEA can be represented with ontologies in a knowledge base. However, there were no similar attempts with FTA diagrams. The later is the main goal of this paper. 4 THE KNOWLEDGE ACQUISTION PROCESS Goals The first goal during the knowledge acquisition process was to identify the components that compose the concept of a material recovery facility and to define their relation. To put it simple, the first goal was to “describe” explicitly the material recovery facility. The second goal was to identify and to analyze as many as possible faults and unwanted events. The third goal was to identify corrective and emergency response actions. Finally, the forth goal was to enrich the explicit description of the facility with the acquired knowledge about the failures and the faults. Achieving the Goals The first goal was achieved by developing an ontology using the ontology editor Protégé. Protégé can edit both frame and OWL-DL based domain ontologies. The development strategy of the domain ontology was top down. It started with the definition of the Materials_Recovery_Facility class and subsequently with the definition of top level classes like: Mrf_Infrastructure, Mrf_Equipment, Mrf_Personnel, Mrf_Input_Waste_Stream, Mrf_Outputs, Mrf_Operational_Problems. The top level class hierarchy is shown in Image 1. Then, for each top level class the subsequent subclasses were defined. This was continued until the desired granularity level was reached. Afterwards, the slots and the attributes of each class and subclass were defined. Through this process more than 40 classes and 160 slots were defined together with more than 30 operational problems. In order to achieve the second goal, a FMEA table and a FTA diagram for each physical object and operational problem were developed respectively. In practice, this was an iterative process during which the knowledge engineer and the manager of the facility were collaborating. The FMEA tables were developed based on the process proposed by [Pillay et al., 2003]. The FTA diagrams were developed based on the process proposed by [Dokas et al., 2006]. The ontology facilitated these processes since a large number of important concepts were already defined during the ontology development. After filling and completing the FMEA tables and FTA diagrams, the facility manager was asked to provide corrective actions for each cause and basic event respectively and thus the third goal was achieved. Finally, in order enrich the ontology with the knowledge from the FMEA tables the process proposed by the [Dittmann et al., 2004] was applied. What was left was to represent the knowledge from FTA diagrams in to the ontology. That will be described in the next paragraph where two general FTA diagrams will be used as example case. These diagrams are displayed in Figure 1. Each diagram is composed of one top event (TE 1 and TE 2 respectively), some logic gates (AND, OR), one intermediate event (IE 1) and some basic events (BE 1, BE 2, BE 3, BE 4). These diagrams are showing that if BE 1 and BE 2 are true then IE 1 is true and if both IE 1 and BE 4 are true then TE 2 is true, or alternative, if either IE 1 or BE 3 is true then TE 1 is true. Image 1. The top level class hierarchy t l Figure1. The FTA diagrams of the example case 5 REPRESENTING FTA DIAGRAM CONCEPTS TO ONTOLOGIES The first task was to identify the key components of the FTA diagrams and based on that to define the corresponding top level classes. In particular, the following classes have been defined: Top_Events, Intermediate_Events, Basic_Events and FTA_Diagrams. The Basic_Events class has been defined to be mutually disjoint from the Intermediate_Events and Basic_Events classes. That is because a Basic Event can not be Intermediate Event neither Top Event in the same or in any other FTA diagram. On the other hand, a Top Event of a FTA diagram can be at the same Intermediate Event in other diagram/s. This allows pointing out potential domino effect/s where a fault can trigger other fault/s. The FTA_Diagrams class has not been selected to be disjoint from the others classes. That is because all the individuals from the other classes are belonging also to FTA_Diagrams class since all these are composing the FTA diagrams. Next task was to define the subclasses of each top level class and to form a class hierarchy tree similar to the one shown in Image 2. Each subclass within a top level class was selected to be disjoint from the others because each individual within a subclass can not be an instance of more that one of these subclasses. In order to describe how FTA (a) (b) Image 2. The class tree Table 1: Subproperties of the FTA diagrams ontology Subproperty Name Domain Range Name of Inverse Subproperty hasBasicEvent MRF_Intermediate_Events FTA_Diagrams Basic_Events isIntermediateEvent hasTopEvent FTA_Diagrams Top_Events isTopEvent hasIntermediateEvent FTA_Diagrams MRF_Intermediate_Events isBasicEvent diagrams are composed by the individuals of the classes the inverse and transitive object properties isComponentOf and hasComponent were defined. In addition, based on these properties a set of inverse subproperties shown in Table 1 and in Image 3 were also defined. A very important goal was to represent into the ontology the structure of the FTA diagrams and specifically to represent the Logic Gate that is associated with the Intermediate Events and the Top Events. To achieve this, two datatype properties were defined. The hasLogicGate with range the Top_Events and MRF_Intermediate_Events classes and the hasStructure with range the Mrf_FTA_Diagrams class. In order to define the Basic Events that are connected with the Intermediate Events a set of conditions was asserted in each subclass within the Intermediate_Events top level class. For example, Image 4 shows the necessary conditions for the Intermediate Event “IE 1” of the FTA diagrams shown in Figure 1(a) and (b). Similar conditions were also asserted in each subclass within the Top_Events class. 6 RESAULTS By using a similar class tree hierarchy together with the properties mentioned above it is possible to represent any FTA diagram in to an OWL DL ontology. For example, the FTA diagram shown in Figure 1(a) can be represented in OWL DL as Figure 2 shows. That was the case in the material recovery facility knowledge acquisition process, where more than 30 FTA diagrams with around 250 Basic Events were represented in to OWL DL ontology. 7 CONCLUSIONS Ontologies can facilitate risk analysis and the dissemination of the knowledge on major risks and accidents. However, the integration of the knowledge about faults, incidents and accidents by the use of ontologies has not been studied thoroughly. This paper proved that FTA diagrams can be represented in to ontologies. In particular it illustrated a way to represent the major components and the relations of FTA diagrams with OWL DL classes and properties. This representation was applied in a knowledge acquisition process for the development of an early warning system in Image 3. Properties and subproperties Image 4. Conditions of the IE_1 class