A FUNCTIONAL ONTOLOGY OF ARTIFACTS

Transcription

1 7. Introduction A FUNCTIONAL ONTOLOGY OF ARTIFACTS The computer systems used in engineering domains include what are called design support systems embodying the sorts of knowledge meaning, in this context, beliefs assumed by experts to be true useful for designing artifacts. For such systems, functional knowledge plays a key role. It is the function of an artifact that is its defining feature; it gives the artifact its identity and it tells us what kind of artifact it is. People create and use artifacts in order to realize their functions because such realization helps them to achieve their goals. Motivated by the importance of function to understanding and using artifacts, we have for some years been developing a functional ontology (Kitamura, et al. 2002; Kitamura 2006). In order to build an ontology it is crucial to maintain a consistent view of reality for example a fhree-dimensionalist or four-dimensionalist view throughout the process of ontology construction. To serve this need for consistency, we have proposed what we call a device-centered view of reality, and it is this which will be explained in this essay. 2. Functions and the Device Ontology 2.1 The device ontology The device ontology corresponds loosely to the three-dimensionalist views defended by many analytic metaphysicians, and is thus contrasted with four-dimensionalist process-centered artifact ontologies in (Forbus 1984). A device processes a certain input to produce a certain output. It is, if one will, the basic atom or unit in the world of artifacts. On the devicecentered view, each artifact is regarded as a composition of devices having as ultimate output some entity that fulfils some need on the part of the users of the artifact. On the process-centered view, in contrast, the artifact "A Functional Ontology of Artifacts" by Riichiro Mizoguchi & Yoshinobu Kitamura, The Monist, vol. 92, no. 3, pp Copyright 2009, THE MONIST, Peru, Illinois

2 388 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA is viewed in terms of the collection of processes that occur in the course of creating this output. The devices of the device ontology can be viewed also as agents, since they play an active role in bringing about a certain output. The process ontology, in contrast, recognizes not agents but rather participants, entities which participate in given phenomena as they occur but without playing any agent role. There is a sense in which it is natural to consider process ontology as more basic than device ontology, since before higher mammals had started to make artifacts, there had existed in the world only natural phenomena. Agency, and design, and deliberate shaping, amounted to something new in history. A device is something built through some deliberate shaping of a certain portion of reality within which certain phenomena of interest occur. This portion of reality is called the environment of the device. The device itself is created (i) by introducing structural components that limit what is produced within this environment to what the device is designed to output, (ii) by connecting these components together to make the outputs intemperate with each other in order to achieve the preset goal. In the present exposition we deal with each device as a black box. Although every device utilizes several natural phenomena to realize its function, this fact will remain implicit. Thus in the context of our present concerns we do not focus on the underlying physics of the device in order to understand how devices work. Rather, our goal is to use the device ontology in order to talk about functions and their realizations in the realm of artifacts in general. 2.2 Advantages of the device ontology The device ontology approach has been dominant in the representation of artifacts for many reasons: 1. This ontology is of very general application; leaving aside the special case of art, every artifact can be viewed in its terms.

3 A FUNCTIONAL ONTOLOGY OF ARTIFACTS In contrast to the process ontology, it allows us to award a central role in our understanding of artifacts to both functions and agents. 3. It brings considerable savings in reasoning, because it allows us to hide the internal details of the workings of devices. 4. It provides a view of artifacts as combinations of devices which is a powerful tool for systems design. The device ontology is not, however, all powerful, since the strategy of viewing devices as black boxes also brings limitations. Certainly one can, where this is required, move to a finer grain in which a parent device is seen as a configuration of several smaller subdevices. But, the smaller devices themselves are then still devices whose internal behavior is hidden. In contexts where one is engaging in creative design, one might need to invent innovative devices based on new combinations of phenomena or on phenomena cross-cutting divisions by which existing devices have been constrained. The process ontology is different in this respect, since it can be used to explain why and how given devices work in a way that addresses directly the phenomena occurring within each device as it realizes its function. 2.3 Constituents of the device ontology The top-level categories of the device ontology include Entity, Role, Structure, Behavior and Function. Entities are continuant objects existing in the world, such as humans, devices, portions of fluid, and so on. During the process in which a device is operating, entities have different Roles, including both functional and structural roles. An example of a functional role is that of Agent. An agent is an actor who performs an action to bring about a desired effect. Subtypes of roles include: Operand, which is the role played by what is processed by the agent; Material, which is the main source of the outputs issuing from the device; and Product, which is the regular output of the device that is made from the Material. The structural role has subtypes such as Input, Output and Component. Component is the role played by any part of a device when it forms a subdevice in its own right. Not all the entities playing the Input role also play a role as Material, though all Materials play an input role.

4 390 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA Similarly, not all entities playing the Output role play a product role, though all Products play an Output role. Structure has subcategories such as Inlet, Outlet and Connection. Each component has inlets and outlets which serve to connect the component with other components. An Input is anything coming into a device through an inlet. An Output is anything going out of a device through an outlet. The Behavior of a device is defined as the "change between input and output" which it brings about. Behavior has two subcategories: Transitive and Intransitive, corresponding broadly to the actions denoted by transitive and intransitive verbs. Transitive Behavior is a matter of the changes in the Operand coming into or going out of the device, Intransitive Behavior of changes in the Agent. Note that "Behavior" here refers not to behavior in general, but only to the behavior of a device. Thus also it does not refer to simulations of devices inside the computer. In the typical case, which is that of numerical simulation, the behavior of a device is understood in terms of the numerical trace of the change in some parameter value over time. The behavior is then seen as the sum of the changes in each parameter value. This account seems both straightforward and convincing. It has its parallel in naive views of behavior which would see walking or running, for example, in terms of the physical trace of the corresponding motion. The device ontology is however close in spirit to another kind of simulation, which is the causal simulation utilized in qualitative reasoning about the behavior of systems as for example in (Kuipers 1986). Such causal simulation differs from numerical simulation in that it tries to simulate the change of an operand between how it exists at the inlet of the device and how it exists at the outlet. That is, in device ontology terms, it tries to simulate the causality involved in the workings of a device. 2.4 Defining 'function' In the world of systems design the three crucial ideas are those of function, behavior, and structure. In order to relate these ideas together, all three need to be articulated in terms of the same ontology of artifacts. On the view defended here, functions are in every case associated with the devices which realize them. More precisely, we see function as reflecting some teleological interpretation of behavior under a goal (Sasajima et al.

5 A FUNCTIONAL ONTOLOGY OF ARTIFACTS ). That is, the change from input to output of a device is interpreted as the exercise of a function when we impute to the device a certain goal. Where a device is assigned multiple functions, this means that it must have several inputs and outputs with distinct input-output pairs selected according to the interests of the modeler. Consider, for example, a boiler which generates hot steam. When water is selected as the input and steam as the output, then the corresponding imputed function is: to transform phase from water to steam; when temperature of water is selected as input and temperature of steam as output, then the function is: to warm. The framework of the device ontology enables us to represent any artifact from a single, consistent point of view and to define its associated functions along these lines. It can be applied recursively through the use of nested device structures whereby an artifact is viewed on successive granular layers, in which the structure at each higher layer specifies the goals to be achieved by the modules at the next lower layer. The topic of function has been extensively investigated in philosophy, but mainly in connection with the functions of biological entities such as lungs or mitochondria. At first glance, it seems to be impossible to define biological function, since organs and cell components do not have goals, while functions, on the view here defended, essentially presuppose a goal. However, we can distinguish what we shall call (1) I-goals, which are the ordinary intended purposes of ordinary agents, and (2) Nl-goals, where intentions are lacking, but where we are forced to talk as if they were present in order to describe an entity from the systems perspective. Where, for example, biological organisms are included as parts in a system which behaves consistently through the collaborative work of its parts, each organism is required to play certain roles definable only in the context of systemic behavior. We then impute Nl-goals to the organism. In the case of artifacts, the two kinds of goals are run together, since any Nl-goal on the side of the artifact is associated with a corresponding intention on the part of the designer. Goals of both types share one characteristic: that there is an endstate that is to be achieved by a certain entity. Whether this endstate is intended by the entity or not is, in our present context, a secondary matter. An I- goal is a goal applied to a thing as a whole, and an Nl-goal is one applied to parts of a certain sort of whole because of their relations to this whole.

6 392 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA Our definition of function above needs only Nl-goals in the latter sense. Biological function can then be defined in the same way as artifact function, by reading 'goal' as meaning 'NI-goal'. The problem of defining function is still not fully solved, however. This is because researchers involved in studying biological function are not happy with the above definition because it seems to appeal to notions of teleology which are for them alien to the scientific approach. 3. Artifact Function versus Biological Function The most salient property of most nonhuman biological organisms is that their existence is independent of human intentions, whereas artifacts are totally dependent thereon. The heart, for example, is in the context of the host organism never used for purposes other than those of pumping and the related purposes necessary for sustaining the life of its host. There is no other organ able to replace it in realizing the pumping function. It owes its origin to its ability to pump, and it will in almost all cases keep on performing this function until the end of the organism's life. An artifactual pump, in contrast, exists in a totally request-based manner. It is designed to realize its pumping function by a designer driven by the requirements of users. More interesting, however, is that artifacts are often used for purposes other than those reflected in the designer's intention. A box can be used as a table (supporting function), as a step (mounting function), as a part of an aisle (guiding function), and so on. Similarly, a pump can be used as a door stop, as ballast, and so on. These essential differences imply two different ontologies for the functions of organisms and of artifacts, respectively. According to (Johansson, et al., 2005), a biological function is a disposition of some part of an organism to act in a certain way ultimately in a way that benefits the whole organism that is its host. The function exists inside the thing which realizes it as a kind of intrinsic property, a property that can be realized in associated processes. On our account of artifact functions, in contrast, it is essential that such functions exist prior to the creation of the products that will realize them in the intentions of the designer. From the artifact perspective, therefore, at least some functions can exist outside of the realizing entity. This is a serious and we believe unbridgeable gap between biological and artifact function. However, we should at the same time not neglect one common feature which different kinds of function share. In both organisms and

7 A FUNCTIONAL ONTOLOGY OF ARTIFACTS 393 artifacts, functions are realized in virtue of intrinsic (physical) features of the realizing entity. According to Johansson et al., a function, as a realizable entity, exists in two ways: as disposition and as exhibition. The former is the function, the latter is the process of functioning. We agree that the function is in every case supported by the structure and properties of the thing. In this respect, there is something essential related to the function that exists inside the thing. At the same time, we also think that the process of realization is essential to what a function is rather than just a process of exhibition. To see this, we need to understand the difference between realization and mere behavior. We believe that the biological definition of function which says that it is a kind of disposition is inadequate for the correct understanding of this difference. Viewed from the engineering point of view, the definition of biological function is very close to that of capability. To see the problem, consider the following case. Imagine we are given a heat exchanger which is composed of a pair of touching pipes in which a hot liquid flows in one pipe and a cold liquid flows in the other. During the flowing process, heat energy is transmitted from the warmer to the cooler flow. This heat transmission between two flows is the behavior of a heat exchanger. This behavior is realized by the touching pipes and the physical law of heat conductance. One might then say: "a heat exchanger has a disposition to transmit heat to reliably achieve a goal specified by the plant." Let us now imagine a situation where people focus on the warmer flow. Then what is realized by the functioning of the heat exchanger is: cooling down the warmer flow, and the relevant function is: to cool. When, on the other hand, people focus on the cooler flow, then what it is realized is: a warming up of the cooler flow, and the relevant function of the heat exchanger is: to warm. The function of a heat exchanger is determined by the intention of its user, and ultimately by the designer of the whole system which has the exchanger as its part. Which flow should be focused on is totally use-dependent. This is an essential factor in the idea of function as understood in engineering domains. Note, however, that the behavior is the same in either case. The very same behavior can play the role of a heater or a cooler according to the context of use, and the very same structure is involved in either case. Structure and behavior are thus independent of viewpoint and context of use, where function is dependent on these. This dependency is essential to artifact functions, which depend

8 394 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA on the intention not only of the designer but also of the user. This implies a radical contrast to biological function, which lies almost completely within the closed world of the bearer of the function. 3.1 Intrinsic versus accidental function The above implies also a further contrast between biological and artifact functions, which has to do with the contrast between intrinsic and accidental functions. Consider the following examples: A door key: Its intrinsic function is to lock and unlock a door. It can however be used also to scratch an itch or to break a seal. The latter are accidental functions. A heart: Its intrinsic function is (inter alia): to pump blood. Accidental functions include the producing of a thumping noise, which can be used for example to serve as a teaching or diagnostic aid. The problem is: do the scratching and thumping functions exist inside the key and heart, respectively? The line of delineation between intrinsic and accidental is a fairly easy one to draw in the case of artifacts. It is determined by the intentions of the designer. All intended functions are intrinsic; all other functions are accidental. In the case of biological functions, in contrast, matters are more difficult. As discussed earlier, each functioning part has its own Nl-goal specified in relation to the relevant including system. If a function of a lower-level part contributes to the behavior of the total system through the contributions of its successive parent systems, then it is intrinsic. Otherwise, it is accidental. In the case of a heart, because no other part of the body exploits the thumping behavior in the sense explained, the corresponding function is accidental. For both intrinsic and accidental functions, their functioning is supported by the same structures in and properties of the bearer. The functions have the same physical base. Therefore, we conclude that intrinsic and accidental functions do not differ in the way they exist inside the bearers of the relevant functions. 4. An Upper Ontology of Function We can now propose an upper ontology of functions for use as a reference ontology intended to serve as the basis for a unified theory of

9 A FUNCTIONAL ONTOLOGY OF ARTIFACTS 395 function. We recently defended our account of function as follows (Kitamura et al. 2006): Function is a role played by a behavior in a specified context. This is to be understood by analogy to: Teacher is a role played by a person in the specified educational context. The heat exchanger similarly plays the roles to warm and to cool according to the focal direction specified in the relevant context. The function of an electric fan is usually considered as either to cool persons or to make them comfortable. According to our suggested account, however, this is not accurate. Let us examine the function of an electric fan step by step. A fan is placed in a room and switched on. It is now realizing the function: to blow. This is the function that is, as it were, closest to the device. The fan is usually pointed towards a person, wind is then blown by the fan. The person feels cooler because the air at body temperature around him is moved away and new, cooler air comes in to replace it. There occurs also evaporation of water from his skin by which heat absorption is promoted, and he thereby becomes cooler. A sequence of physical phenomena of this sort must occur after the wind has been blown by the fan if a cooling function is to be realized. The whole is represented in terms of functions as follows: to blow wind -» to blow wind at a person» to move air away» to promote evaporation» to generate heat absorption The example shows how, when some device function is executed, many other phenomena occur consecutively leading to certain desirable effects at the target. Another example is an analog clock. Here the function is usually understood as: to show the time. This, again, is not accurate. The function of an analog clock is just: to rotate the long and short hands in a certain regular manner. Time is read into this rotation by humans. Human interpretation here plays a key role. As these two examples show, there exist several types of function in the world, and we are afraid that this important fact has not been adequately recognized thus far. The function performed by a concrete device itself is just one of a number of different types of function, which can be

10 396 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA organized in terms of a hierarchy of function specifications along the lines shown in Figure 1. Figure 1: A hierarchy of different uses of 'function' (slightly modified from Kitamura et al. 2007). Here function specifications are divided into three types: requirement functions, effect functions and property functions. An effect function is one whose realization brings about some effect on the target. Explanations are shown in the table below and on the following pages. Function Effect function Requirement function Property function Explanation Function specified in terms of the effects its execution brings about on the target. Function specified in terms of some requirement. Examples are found in the field of functional materials, in which elasticity, conductivity, etc. are dealt with under the heading 'functions of materials'.

11 A FUNCTIONAL ONTOLOGY OF ARTIFACTS 397 Function System boundary function Environmental function Device function Characteristic function Disposition function Way function Genuine effect function State-operational function Process-operational function Explanation Functions under this heading are: to import and to export. Both concern movement of things across the boundary between the the device and its environment. Function whose execution includes effects caused by phenomena occurring in the environment (as in the case of the electric fan). Function that is properly attributed to its bearer (thus excluding system boundary and environmental functions). Function specified on the basis of the characteristics of the device rather than on input/output relations. Function specified on the basis of the disposition of the device. Function specified in terms of manner of realization (for example: the function to weld, where it is specified that two sheets of metal should be joined together through fusion). Function specified in terms of what is to be achieved rather than of the manner of its achievement (contrast: to join with: to weld). Function specified in terms of state changes within the device assuming no change in the device structure while functioning occurs. Function specified in terms of the processes in which the device is involved rather than in terms of changes in objects.

12 398 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA Function Explanation Inter-device function Meta-function User-action enabling function Environmental-interpretive function Physical environmental function Function realized by one device in relation to another (for example: the pushing function of a rod within a cam). Function to another function (for example: to enable). Function specified in terms of what users can do (for example: the function of a conference room). Function realized through human interpretation (for example: the function of a clock to indicate time). Function realized with the help of of physical phenomena causally influencing the final state. Table: Explanations of the different ways functions are specified, as represented in Figure 1. Although requirement functions have the same structure as effect functions, they are classified separately because of their different origins. Property functions include biological functions and characteristic functions, by which we mean the functions associated with what are called functional materials in materials science, such as high conductivity, heat resistance, and so on. Effect functions are divided into the three subclasses of system boundary function, device function, and environmental function. Environmental functions are further divided into: physical environmental functions, such as the cooling function of evaporation; environmental interpretive functions, such as showing the time; and user-action enabling functions, such as the function of a meeting room to enable a meeting. Device functions consist of genuine effect functions and way functions. Genuine effect functions consist of state-operational functions and process-operational functions. The latter have what we call meta-functions such as to enable or to prevent, which have an effect on processes which are the realizations of other functions. At the basis of this

13 A FUNCTIONAL ONTOLOGY OF ARTIFACTS 399 classification are what we can think of as the 'normal' functions executed by artifacts satisfying our definition of 'device' within the framework of the device ontology. Examples of such normal functions include to separate, to join, to transform, and so on. They are applied in explanations of how engineering artifacts work and of how they are configured through connection of functional modules. Normal functions as we understand them are implementation-independent. That is, they are defined by the fact that their execution is identical to what users want to enjoy who are uninterested in how a function is realized. It is obvious that there exist multiple ways of realizing any given normal function. In theory, each new engineering design signifies the invention of some innovative way of realizing one or other of these normal functions. Typically, the job of engineering designers is to select one out of the candidate ways to realize the functions they are dealing with. There are many types of functions that have not as yet received proper recognition. This is so especially in regard to the family of what we are calling way functions such as: to cut, to wash, and so on. To cut implies the use of a sharp tool and to wash implies the use of a liquid. Such functions are associated in obvious ways with specific kinds of harms, the failure to pay attention to which has been a chief cause preventing smooth reuse of functional information across domains, since how such functions are to be realized is highly sensitive to domain-specific (physical) features of the relevant bearers. The definitions of normal functions, in contrast, because they are free from this domain dependence, are sufficiently generic to capture essential functional structure. We have formalized definitions of 89 distinct functions of the normal function type in the Web Ontology Language (OWL) (W3C 2004, Ookubo et al. 2007, Kitamura et al. forthcoming). Figure 2 shows some normal functions and associated ways of functioning together with an example of their use in functional decomposition of a type which has proved itself useful for many real-world purposes (and which has indeed been applied by us in close collaboration with three companies over several years). Using this catalog of 89 functions we have developed a tool called SOFAST for representing the functional structure of any artifact that has been deployed in industry since The National Institute of Standards and Technology (NIST) in the USA has independently developed a list of functional terms called the Functional Basis (Hirtz, et al., 2002). This includes 52 functional terms, though without any underlying ontology. We

14 400 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA Functional Concept Ontology (is-a hierarchy of generic functions) transform electricity to heat Example of is-a hierarchy of generic ways of function achievement for removing entity pays for removing entity Figure 2. Relations among functions. ] Physical way Example of a function decomposition tree of a coffee maker (part-ofh ierarchy) have investigated the Functional Basis with the intention of aligning the two sets of terms, and found more than 80% of its terms have correspondents in our own ontology. Considering the independence of these two bodies of research, this high correspondence rate provides a strong confirmation that we are both on the right track, and that our functional ontology successfully captures the essential functional concepts employed in engineering. At the same time we found some room for improvement in the Functional Basis, turning in part on its lack of ontological analyses. 5. Concluding Remarks We have discussed the ontology of functions from an engineering point of view, and included a comparison of biological and artifact functions. The major difference between the two lies in the fact that artifact functions are more sensitive to user requirements and to the context of use than are biological functions. The function to be realized by an artifact is essentially realization-independent; nobody, not even the designer, knows how it is going to be realized before the design is created and implemented in some device. At the same time, even here the realized function is essentially supported by the structure and properties of the function's bearer (i.e., by the device). Thus as in the biology case so also

15 A FUNCTIONAL ONTOLOGY OF ARTIFACTS 401 in the case of artifactual functions, the physical basis in the bearer is essential to the function. The authors thus believe that a unified theory can be developed that would cover both kinds of function. The key idea would be to identify finer-grained functional types reflecting various modes of functional entities such as when a requirement is given as a functional specification, when design is being performed, after design has been finished, when it has been manufactured, when it is used, etc. before designing the unified theory which is expected to explain each of the above. Functions of existing devices which are in the mode of after-manufacturing would have function as disposition. Another idea which could be consolidated with the one mentioned above is introduction of "evolution" of function along with that of biological entities. We believe that it is worth investigating the possibility of producing a unified theory along these lines, and of refining the upper functional ontology for this purpose. Such a unifying theory should involve a more careful examination of what exists inside the two types of bearers of functions and of how the two sorts of realization differ. Osaka University Riichiro Mizoguchi and Yoshinobu Kitamura REFERENCES Forbus, K.D "Qualitative Process Theory," Artificial Intelligence, 24: Hirtz, J., R. Stone, D. McAdams, S. Szykman, and K. Wood "A Functional Basis for Engineering Design: Reconciling and Evolving Previous Efforts," Research in Engineering Design 13 (2): Johansson, I., B. Smith, K. Munn, N. Tsikolia, K. Eisner, D. Ernst, and D. Siebert "Functional Anatomy: ATaxonomic Proposal," Acta Biotheoretica, 53 (3): Kitamura, Y., T. Sano, K. Namba, and R. Mizoguchi "A Functional Concept Ontology and Its Application to Automatic Identification of Functional Structures," Advanced Engineering Informatics, 16(2): Kitamura, Y. 2006, "Roles of Ontologies of Engineering Artifacts for Design Knowledge Modeling," In Proceedings of the 5th International Seminar and Workshop on Engineering Design in Integrated Product Development (EDIProD 2006), Gronow, Poland, Kitamura, Y., Y. Koji, and R. Mizoguchi "An Ontological Model of Device Function: Industrial Deployment and Lessons Learned," Applied Ontology (Special Issue on "Formal Ontology Meets Industry"), 1 (3-4): Kitamura, Y., S. Takafuji, and R. Mizoguchi "Toward A Reference Ontology for Functional Knowledge Interoperability," Proceedings of the ASME 2007 Internation-

16 402 RIICHIRO MIZOGUCHI & YOSHINOBU KITAMURA al Design Engineering Technical Conferences and of the Computers and Information in Engineering Conference (IDETC/CIE 2007), Las Vegas, Nevada. Kuipers, B "Qualitative Simulation," Artificial Intelligence, 29: Ookubo, M., Y. Koji, M. Sasajima, Y. Kitamura and R. Mizoguchi "Towards Interoperability between Functional Taxonomies using an Ontology-based Mapping," Proceedings of the 16th International Conference on Engineering Design (ICED 2007), Paris, France, August 2007, CD-ROM, No. 154 (Abstract: ). Sasajima, M., Y. Kitamura, M. Ikeda and R. Mizoguchi "FBRL: A Function and Behavior Representation Language," Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI-95), W3C OWL Web Ontology Language Overview, (accessed November 27,2008). Yoshinobu K., S. Segawa, M. Sasajima, S. Tarumi, and R. Mizoguchi (forthcoming). "Deep Semantic Mapping between Functional Taxonomies for Interoperable Semantic Search," Proceedings of the 3rd Asian Semantic Web Conference, Bangkok, Thailand, 2008.