Design of a manufacturing knowledge model

A manufacturing model is an information and knowledge model that describes the manufacturing capability of a particular organization. This work contributes to the area of information and knowledge structure to support manufacturing decisions. The structures of the manufacturing model have been defined to achieve suitable access to, and maintenance of, the manufacturing knowledge. Emphasis has been made on investigating a suitable manufacturing model structure in order to readily access manufacturing knowledge related to process planning activities. The aim of this research was to design a manufacturing knowledge model (MKM) and demonstrate its functionality through experimental software. This paper presents requirements for a new MKM, proposes its structure and describes its design.

Keywords: Manufacturing model; Knowledge model; Information and knowledge structures; Decision support; Process planning

Manufacturing companies utilize large amounts of knowledge to manufacture products and apply manufacturing knowledge to develop products according to customer requirements and to offer competitive prices. Over time, the combination of new technologies, product innovation and experience of workers has improved the manufacturing knowledge used to produce articles and, as a consequence, this collective knowledge has expanded. To improve product development decisions and to obtain a competitive advantage, an important aim for manufacturing companies is to retain, transfer and improve their own manufacturing knowledge (Beckett [2]).

Manufacturing is a set of related tasks including product design, material selection, planning, production, etc. Process planning involves selecting which processes are to be used to convert a commodity from its initial form to a determined final form (Chang et al. [6]) as well as a suitable sequence of such processes. The research presented here applies to process planning of machining holes in a workpiece.

The concept of information and knowledge modelling is an accepted manner of integrating the information pertaining to a particular manufacturing plant and it is recognized as an area of research where effective results can have wide-ranging implications for improving the decision-making process. Significant advances have been made in manufacturing information modelling to support decisions throughout the product development process using the concepts of (a) the product model, which captures product characteristics, and (b) the manufacturing model, which stores manufacturing information. Even though implementation of these models is mainly focused on supporting the decision-making process, there is a need to develop flexible computerized systems that can be readily maintained with up-to-date information gathered by a manufacturing model (Young [31]).

Within manufacturing companies, knowledge based systems (KBSs) are used in product development processes to support manufacturing and design decisions, and are considered as decision-support systems. These software systems are important tools with which to obtain a competitive advantage and leverage using what the company knows, contained in its manufacturing model. Due to the substantial volume of knowledge generated in manufacturing and design stages, there is a need for a new approach to structure and handle knowledge that enables readily maintenance (Rezayat [23]), assuring the long-term use of these software systems (Young et al. [32]). The majority of current methods for maintaining manufacturing knowledge of a KBS have one or more of the following shortcomings: (a) they are unable to deal with various types of knowledge; (b) they require huge time and effort to constantly update knowledge; (c) they exhibit inadequacy in capturing new and valuable knowledge (Guerra and Young [11]). This research deals with a manufacturing knowledge model (MKM) able to overcome these shortcomings when integrated to an experimental KBS.

It is important to have an understanding of information and knowledge related to the manufacturing facility under study (Young [11]). A manufacturing facility has processes and resources; its components are represented by its model and categorized throughout a structure. Comprehension of the differences between information and knowledge related to processes and resources is required to define data, information and knowledge, as well as to differentiate the types of knowledge used.

Data relate solely to words or numbers whose meaning is dependent on the context in which they are used. Information is data structured to provide a meaning within a given context (Harding [12], Guerra [10]). Knowledge has many possible interpretations; knowledge exists when a user within a specific context recognizes useful data relationships or when a user is able to derive those relationships from raw, but structured, information (Guerra and Young [11]).

Guerra and Young ([11]) emphasized the use of some types of knowledge in a manufacturing knowledge structure, related to different areas within a manufacturing organization. Three main types of knowledge were used in that research work: explicit, tacit and implicit knowledge. Explicit knowledge is objective and rational, and is captured in storable documents such as texts, tables, formulas, diagrams, product specifications, graphs, etc. Tacit knowledge is subjective and cannot be articulated (Nonaka and Takeuchi [22], Nickols [21], Mills and Goossenaerts [18], Zheng et al. [34]) but it refers to well-tested knowledge and follows a format (Guerra [10]). Implicit knowledge is the type of knowledge that has not yet been articulated; it is implied by or inferred from observable behaviour or performance of another person (Nickols [21]).

The broad research work reported here aimed to build a new manufacturing model able to store and manage various types of knowledge (explicit, tacit and implicit) by means of using a knowledge model to comprise both information and knowledge related to a manufacturing facility.

This paper is structured as follows. Section 2 presents the requirements for the design of a MKM, which is explained in section 3 using object-oriented (O-O) concepts. section 4 describes the experimental software developed to support the research ideas, while section 5 contains a discussion of results, conclusions and further work.

The research presented in this paper relates to modelling information and knowledge of a manufacturing facility that performs machining operations to produce holes. Prior to design, a manufacturing model to describe its components is required to collect what the organization 'knows' and then categorize it utilizing a suitable representation. This information and knowledge collection can be used as a source and repository to support decisions at a later stage by the use of a KBS.

Several researchers have defined the different types of knowledge that humans use to make decisions. The work reported by Nonaka and Takeuchi ([22]) was used as a basis for this research, although others were also taken into consideration (Mahe and Rieu [15], Rodhain [24], Malhotra [16], Mascitelli [17], Nickols [21], Beckett [2], Mills and Goossenaerts [18], Swap et al. [29], Zheng et al. [34], Ackerman et al. [1], Chung et al. [7]). Machining knowledge types used in this research work are classified as explicit, tacit and implicit. In addition, these knowledge types are further sub-classified to enable representation of processes and resource knowledge structures. Explicit knowledge can be represented through procedures, tables and graphs. The representation of tacit knowledge may include patterns, storytelling, video clips and sketches. Implicit knowledge is represented by text (Guerra and Young [11]). By capturing the right knowledge in the right format, valuable knowledge loss is avoided (Rezayat [23], Kreiner [13]).

By analysing the diversity of resources and processes of a manufacturing facility, it can be established that there is a lot of facility knowledge related to different subjects at different levels of detail. Any manufacturing information or knowledge gathered can be related to either a process or a resource of the facility.

Process information could include, for example, the identification of current manufacturing processes and attributes related to each process identified. From such process information it is possible to gain a general overview about the process itself and to understand which processes can be performed by that particular facility. In contrast, to know how each manufacturing process can be used to produce a manufactured feature is considered process knowledge.

Resource information is structured data that uses different attributes to identify and describe a specific resource. Resource knowledge is structured information with added detail; it describes how the resource can be used related to previous experiences of hole production (in the scope of this research).

Facility knowledge is an important part of the intended knowledge model because it contains all the process and resource knowledge identified in the manufacturing facility. Therefore, it is necessary to define a structure that allows the access and storage of the wide range of facility knowledge. To define these knowledge structures it is necessary to explain what process and resource knowledge the manufacturing facility has and how they can be represented. Some examples of collected knowledge are presented to clarify the definitions of knowledge representation used in this work.

Knowledge related to manufacturing processes is structured using different types of knowledge representations. The authors make use of a variety of knowledge representations in this work. Through these representations, a wide range of process knowledge that technicians already possess to produce a hole is shown.

Diameter tolerance, positional tolerance, roundness and surface finish form, are in general, the main information used to select a milling process to manufacture a round hole. However, an important point is to define which knowledge representation is appropriate to organize this information. In this case a table can be used to organize process knowledge.

Table 1 shows the dimensional parameters used in this particular facility to decide which milling process is suitable to produce a round hole. This explicit process knowledge was obtained according to the technician's experience working within a CNC shop. This example shows how a technician is able to gain knowledge through their work experience and how it can be structured in order to develop a decision. Nevertheless, a table can contain other types of knowledge representations inside its cells such as procedures, storytelling, video clips, etc.

Table 1. Explicit knowledge representation to select a milling process.

Different methods can be used to determine the spindle speed to drill a hole. For example, graphs can indicate recommended revolutions per minute (rpm) for various materials for different drill diameters, e.g. the graphs found in Titex or any other reference manual (Titex [30]). This knowledge was obtained as a result of previous experience in drilling different types of materials. These values are, however, only recommendations and should be adjusted according to the working conditions and properties of the material being worked.

This kind of knowledge representation is recommended to store process knowledge when a technician says the common phrase 'I know how to do this but I can't explain it' (Nonaka and Takeuchi [22]). For some people it is extremely difficult to organize their know-how using explicit representations such as graphs. In such cases it is necessary to work with knowledge elicitation and tacit knowledge representation techniques to make available new knowledge for other people within the same manufacturing organization.

The manufacture of a tapered hole in the manual shop studied required the creation of a dedicated fixture to hold the workpiece according to a technician's know-how, as depicted by figure 1. A procedure to use this new fixture was defined by the technicians as follows:

• a. locate fixture plate on lathe plate using grips
• b. align fixture plate at lathe centre line according to pilot hole using dial indicator
• c. locate workpiece on fixture plate using grips and pin
• d. remove pin and turn tapered hole
• e. turn next workpiece repeating from third step.

Graph: Figure 1. Sketch representation.

Table 2 depicts a pattern to produce tight tolerance holes using a jig-boring machine. This tacit knowledge was obtained according to the technician's know-how working within two manufacturing shops, one manual and one with CNC equipment.

Table 2. Pattern for tacit knowledge representation.

Explicit and tacit knowledge can be used in a different manner to support process planning decisions (Sormaz and Khoshnevis 1997). A specific answer is obtained each time a graph is consulted to confirm a value; however, using the pattern tacit knowledge representation depicted in table 2, different useful ideas can be obtained to produce tight-tolerance holes using a jig-boring machine. The combination of text explanation and format provides a valuable tacit knowledge representation and wider knowledge context to support machining decisions.

Storytelling title. Useful ideas to reduce burring problems producing a hole in aluminium.

Storytelling context. Last June 2000 excessive burring occurred when trying to produce a hole (8 ± 0.1 mm diameter and 25 mm length). All the conditions were according to current procedures and handbooks. The problem was reduced by 20% by using paraffin as a lubricant.

Storytelling and patterns provide different layouts to store process knowledge, but the manner to use the stored knowledge depends on each particular case. Storytelling could be suitable when the know-how could be transferred in a text manner without too much detail (Swap et al. [29]). It can be useful to write down personal notes related to day-to-day problems in order to identify the main problem and how it was solved. On the other hand, patterns provide more knowledge related to a particular problem. Furthermore, different storytellings can be used to produce a pattern. Though storytelling and patterns use different methods to store technicians' experience, both can be used to store tacit knowledge and support machining decisions.

In this work, the definition of implicit knowledge includes only new technology knowledge, not well-tested knowledge as tacit knowledge involves. Implicit knowledge is only a text message with no format. There are, for example, different experiences applied to produce a round hole where implicit knowledge is used to organize explicit and tacit knowledge. For example, milling knowledge is know-how of the drilling process performed in a milling machine. On the other hand, turning knowledge is know-how of the drilling process but performed on a lathe. However, the experience of turning knowledge can be applied to structure milling knowledge. The new techniques and technologies in workpiece alignment during turning operations can sometimes be used to structure milling knowledge and support milling decisions. The definitions of Nickols ([21]) for implicit knowledge apply due to the characteristic of being inferred by a behaviour observed and then extrapolated to a new manufacturing problem or technology in use.

An additional example of implicit knowledge is know-how about ultrasonically assisted drilling in aluminium alloys that can help to reduce, and under some conditions completely eliminate, the burring problem (Babitsky et al. [3]). A concise explanation related to this new technology could be useful to support milling and turning decisions. This new technology concept has useful implicit knowledge that can be applied to build new milling and turning knowledge.

Sometimes technicians apply only their own past experience, personal notes or handbooks to obtain an answer. Whilst it is not wrong to work individually in this manner, it is important to obtain a method to structure the whole organizational process knowledge contained in such personals notes and handbooks to support future decisions. Three advantages can be obtained by structuring process knowledge: (a) experienced technicians can increase the value added in solving problems combining different experiences; (b) novice operators can learn through following the knowledge already structured; and (c) a wide range of new knowledge can be accessed at different levels of a manufacturing company using different representations.

According to the scope of this work, resource knowledge exemplified in this section relates only to manufacturing operations to machine a hole; manufacturing resources for this particular case involve drilling tools and machinery to perform manufacturing operations.

An example of resource knowledge used to support decisions is related to a twist drill before and after the drilling process is performed. The knowledge to select the twist drill to produce a hole in a workpiece is not overly complex. The hole diameter and its length could be enough to select a suitable twist drill to manufacture such a hole. Conventional KBSs name this kind of knowledge production rules that are declared in the form IF premises THEN conclusions (Shehab and Abdalla [27]). For example, an IF–THEN rule can be abstracted from a sketch that appears in Titex ([30]) in the following manner. If the drill tool has a double angle point then it is recommended for cast iron and abrasive materials. In the production of the workpiece, there are items of knowledge related to the twist drill that are more difficult to represent in the form of production rules, for instance, the steps to re-sharpen a twist drill could be better represented via a procedure.

Many industrial and research groups have developed the concept of a manufacturing model as a tool for managing manufacturing information (e.g. Molina et al. [20], Harding [12], Giachetti [9], Molina and Bell [19], Zhao et al. [33], Dorador [8], Liu and Young [14], Rodriguez and Al-Ashaab [25]).

The ideas presented in this work extend Molina and Bell's ([19]) approach with a significant contribution to knowledge structures for both processes and resources that were not emphasized in that work. Although Molina and Bell proposed a process and resource classification, new key relationships between information and knowledge related to processes and resources in a manufacturing facility were identified throughout the present research work. These new relationships made necessary the deployment and structuring of a facility knowledge category, a process knowledge class and a resource knowledge class.

The unified modelling language (UML) is a notation for representing software blueprints designed for a broad range of applications. It provides software construction utilities for a broad range of systems and activities (e.g. real-time systems, distributed systems, analysis, system design, deployment) (Bruegge and Dutoit [5]). Visualizing, specifying, constructing and documenting O-O systems are exactly the purpose of the UML (Rumbaugh [26], Booch et al. [4]).

Figure 2 shows a UML top-level class diagram, emphasizing the area of the authors' contribution related to Molina and Bell's work.

Graph: Figure 2. Manufacturing model top-level structure (adapted from Molina and Bell [19]).

The research reported in this work identifies three new classifications in the new MKM to readily access and store the manufacturing information and knowledge. These classifications are: (a) processes knowledge, (b) resources knowledge and (c) facility knowledge.

A key issue when defining the knowledge structure was how to determine and categorize knowledge in a manufacturing facility. Different knowledge structures were explored, for example, knowledge as a resource and process attribute was considered. However, because of the large number and different types of knowledge, this option is very difficult to implement. The quest to find an optimal structure continued and several potential options were analysed until it was found that there are two superclasses that provide a wide range of facility knowledge categorization. The first superclass was named facility knowledge, i.e. the knowledge related to processes and resources is categorized. An additional class, 'types of knowledge', was used to categorize current knowledge types. Combinations of these superclasses made the manufacturing facility categorization possible.

The whole manufacturing knowledge is organized in a facility knowledge superclass that is divided into process knowledge and resource knowledge. The facility knowledge superclass can include different manufacturing knowledge such as: grinding, EDM, lapping, casting, etc. However, two main classes are considered in process knowledge, milling knowledge and turning knowledge according to the research scope. The whole process knowledge is organized in the milling and turning subclasses. In a similar manner three main classes of the resource knowledge are considered: machinery, tool and material.

Creation of knowledge structures requires understanding of the knowledge subject and oriented concepts. The knowledge subject is performed by the know-how required to support particular decisions. The knowledge categorization follows O-O concepts (Rumbaugh [26], Booch et al. [4]). For example, drilling, reaming and boring have been defined as knowledge subjects but can also be considered as milling subprocesses and, as a consequence, can inherit all the process knowledge superclass attributes. Drilling, reaming and boring processes have been considered as milling processes according to the ISO 14649-11 standard. However, these milling processes can be performed on a lathe, and then the whole knowledge related to drilling, reaming and boring processes performed on a lathe is considered in the turning knowledge superclass. The creation of a proper structure for these processes requires an understanding of how they are related and how their relationships can be reproduced in an O-O programming language.

According to the different types of knowledge defined previously, this section discusses the organization of these knowledge types according to O-O concepts.

A superclass is defined to organize different types of knowledge and use this knowledge to access the facility knowledge. A superclass named types of knowledge is defined to organize the current knowledge types in a manufacturing facility. Explicit, tacit and implicit knowledge are considered subclasses of this superclass. The explicit knowledge described is divided into table, graph and procedure subclasses. In a similar manner, the tacit knowledge can be divided into sketch, pattern, video clip and storytelling. Implicit knowledge is considered in the implicit knowledge class. Figure 3 shows an explicit, tacit, implicit type of knowledge structure to represent facility knowledge. It is important to define suitable attributes according to the type of knowledge definition and the particular knowledge representation in order to capture the right knowledge in the right format. For example, the 'type of knowledge' superclass has attributes such as: name, type of representation, identification (ID), number of changes, modified date. These attributes are characteristics that facilitate the management of each facility knowledge instance created. In addition, each type of knowledge representation subclass has particular attributes as can be observed in figure 3.

Graph: Figure 3. Explicit, tacit and implicit types of knowledge structure.

After several versions of the structure, the final MKM structure depicted in figure 4 was chosen since it was better at organizing information and knowledge in a manufacturing facility. Although it shows specific resources and processes for this machine shop, the parent classes can be used to describe another facility under study. This MKM structure can capture a significant amount of information and knowledge related to processes and resources in a manufacturing facility.

Graph: Figure 4. Information and knowledge structures in the MKM.

It is difficult to separate information from knowledge in a manufacturing facility. In spite of this, according to the information and knowledge concepts previously discussed, it was possible to create the information structure presented to capture process information. A new knowledge structure was created, allowing a wide range of process knowledge to be captured in a manufacturing facility. This section discusses the organization of the process knowledge using the process knowledge structure depicted in figure 5.

Graph: Figure 5. Relationships among process knowledge classes and knowledge representations.

There are large amounts of complex process knowledge in a manufacturing facility. The hole-making process knowledge was taken as an example. To support the hole-making process, milling and turning knowledge were identified as types of hole-making process knowledge. The knowledge structure obtained allows the capture of the process knowledge.

After gathering available information and knowledge, an additional issue is to define the right process knowledge attributes to enable suitable process knowledge identification and classification. The attributes such as name, classification and application have been defined to identify the facility knowledge instances. In this case, according to figure 4, the knowledge name is process knowledge and the classification could be milling or turning knowledge. In addition, the definition of knowledge application depends on the drilling, reaming or boring knowledge subject. As a consequence, each instance of the classes process selection, drilling process, reaming process and boring process can be represented by explicit, tacit or implicit knowledge. According to the arrows depicted in figure 5 it is possible to identify the connection of the different process knowledge instances considered.

The turning knowledge found in this facility is an instance of turning knowledge class and is also an instance of the sketch class. In a similar manner the process selection knowledge represented is an instance of process selection knowledge class and is also an instance of a table because it contains the knowledge of table 1. The majority of the knowledge discussed is an instance of the drilling process knowledge class. For example, explicit knowledge represented in table 1 is an instance of drilling process knowledge class and an additional instance of table class. The procedure to manufacture the round hole is an instance of the drilling process knowledge class and is also an instance of the procedure class. The explicit knowledge represented is an instance of the drilling process knowledge class and is an instance of the graph class. The tacit knowledge represented by the storytelling is an instance of drilling process knowledge class and is also an instance of the storytelling class. Hence, the knowledge structure elements and their interactions are identified.

The UML tool is useful in the creation of new structures detailing attributes for defined knowledge classes. An experimental software system should be able to implement research ideas and demonstrate the functionality of the MKM using the O-O database Object Store® and the Visual C++ programming environment. This software system is intended to select the best process to machine a round hole based on IF–THEN rules, where the decision criteria involves dimensions and tolerances according to table 1; additionally, it will serve as a knowledge repository for the manufacturing shop, thus allowing sharing, reuse and updating of such knowledge.

The relationships between knowledge classes of the MKM indicate, for example, the type of knowledge utilized to represent either process or resource knowledge. This experimental software system was designed to allow definition of these interactions after being populated with information and knowledge instances.

Experiments were performed focusing on exploration of the MKM structure to test how this model can support process planning decisions and storage, updating and sharing of manufacturing knowledge. The first experiment covered a process selection to machine a round hole while displaying all available knowledge related to the selected machining process.

Process-selection-knowledge class and drilling-process-knowledge were associated to specify what the organization knows relating to the drilling process. Consequently, process-selection-knowledge instances uses table as a type of knowledge representation and this table collects the decision criteria organized in table 1; similarly, the drilling-process-knowledge instance uses a procedure and implicit knowledge representation to explain how the process is performed.

At this stage the round hole instances were stored, and the information and knowledge related to the process was linked as well; the next step is to show how a round hole selected applies the process information and knowledge stored.

Figure 6 depicts the dialog boxes in which the product instances are shown. Additionally, the dialog boxes also show the features of the round hole instances stored in the MKM. For example, two existing round hole instances can be observed: (a) through end round hole and (b) tapered round hole.

Graph: Figure 6. Process planning for a round hole example.

The experimental software developed can show round hole attributes by clicking on the particular round hole instance. The detailed design for the round hole can be shown by clicking on the 'show me design' button. The experimental software supports process and tool selection for the round hole selected. When pressing the buttons 'process selection' and 'tool selection', the experimental software runs a computational IF–THEN rule in order to select the right milling process; this rule criteria takes into account the diameter and positional tolerance, the roundness and roughness required by the user and compares these data with the stored information in order to propose the best choice.

Process planning support decisions using current knowledge are in two steps, as exemplified with the through end round hole instance. Firstly, the process selection showing the information and knowledge for the process recommended is presented. Secondly, the tool selection showing the information and knowledge for the resource selected is presented.

Figure 7 depicts the dialog boxes in which the process recommended for the selected through end round hole is shown. According to the hole attributes, the system suggests a drilling process.

Graph: Figure 7. Decision support using information and knowledge stored within the MKM.

It is important to emphasize that by clicking on the process recommended, the process information and the process knowledge will be displayed in separate dialog boxes. By clicking on the 'Process selection' button, the software can provide a suitable machining process to produce a round hole. Additionally, the software can provide current information and knowledge related to the process recommended (figure 8). The seven-step procedure shown describes a technique to drill based on the technician's experience.

Graph: Figure 8. Process knowledge related to a machining process selected.

Some other examples were conducted with this software system. The second experiment dealt with a tapered round hole instance used to support process planning decisions using new information and knowledge. The third experiment aimed to share knowledge related to processes and resources when planning machining processes for both a tapered and a round hole, using both current and new knowledge. The fourth test explored the consequences that a change in the rule-based criteria has in the event of a change in technology.

A new manufacturing model was designed to ensure management and storage of different types of knowledge representation using knowledge structures. The MKM was focused on the improvement of decision-support systems; it provided structured information and knowledge to store facility information and knowledge related to processes and resources. Through this work it was confirmed that an effective MKM structure should represent not only information related to processes and resources, but also knowledge associated with the processes and resources with the intention of modelling all of the manufacturing facility components and capabilities.

This research utilized types of knowledge with the purpose of characterizing manufacturing knowledge. The design and definition of the information and knowledge structures required an in-depth analysis of information and knowledge modelling. The use of tools and techniques for information and knowledge modelling was an effective support in the design of the experimental system. UML provided a consistent notation for the creation and representation of the classes and detailed attributes for the design of the MKM structures and the development of the experimental software system.

The experimental system designed and implemented here has proved to be adequate for exploring the research ideas discussed in this work. A variety of manufacturing cases were used to test and validate the research concepts.

Manufacturing information and knowledge related to processes and resources to support process planning decisions were used to explore the MKM structure. A machining shop at Loughborough University was used for practical applications of machining holes in a workpiece, as an example of manufacturing process. The MKM was validated at station level but it would benefit from being tested at higher manufacturing levels, i.e. enterprise or factory level. Furthermore, it would be useful to investigate how applicable the MKM structure is to support other manufacturing decisions regarding resource selection, material selection, etc., and to use other types of knowledge representation.

The knowledge, captured through the model structures and their relationships, could be easily updated by applying a knowledge maintenance method. Throughout this broad research work, a knowledge maintenance method was defined and the results obtained will be reported in a future paper.

UML structures were identified as static structures, raising the need to develop new dynamic structures with an aim to provide maintenance to the knowledge repository in a timely fashion. Dynamic structures are currently under study, as are compatible methods for updating both information and knowledge.

The authors intend to investigate how this MKM could be used by other decision-support systems based on knowledge, and to explore the differences between tacit and implicit knowledge to improve definitions and contribute to state-of-the-art systems.

We acknowledge support from Loughborough University and supplementary support from CONACYT, México. Additional support for this work was provided by Tecnológico de Monterrey through the Research Chair in Autotronics and by the IBM SUR GRANT.