MANUFACTURING SYSTEM RE-ENGINEERING USING COMPLEXITY

By Shih Bun Yu and Janet Efstathiou

Abstract

This paper describes a new approach, structural complexity, to assess manufacturing systems. Structural complexity is the information required to monitor manufacturing systems with specific focus on the structure and the effect of inter-dependency of components within the system on the performance.

We demonstrate the value of the new innovative measure when applied to system design and re-engineering. Manufacturing system re-engineering can take many forms. Four identified fields of re-engineering techniques are discussed, namely layout re-structuring, buffer size adjustment, modifying the parameters of individual station and rework cell re-configuration.

One of the major contributions is to summarise different re-engineering approaches into two categories which are flexible and lean manufacturing according to the value of structural complexity. In addition, the result reveals some useful insights for better modifications, particularly in rework cells and buffer installed systems.

1. Introduction

Flexible manufacturing design is undoubtedly versatile and can handle multi-product and multi-processing in an integrated system. However, a huge cost must be paid to construct and manage such a flexible network. For this reason, there is another trend to promote lean manufacturing design since the last decade. Streamlining and simplifying system design not only saves money for advanced material handling tools and spaces for flexibility, but also reduces the management difficulty to a minimum by lowering the complexity [Womack and Jones 2005].

This paper aims to formalise and quantify these intuitive observations and establishes the relationship between complexity, cost and other important performance indicators. The remainder of this paper is organised as follows. Section 2 provides the literature background on complexity measure. Section 3 discusses the recent research direction of manufacturing system re-engineering. Section 4 briefly illustrates the new structural complexity measures used in this paper. In Section 5, the results of four types of re-engineering are summarised. Section 6 summaries this paper by and classifying re-engineering strategies into two major design trends, which are flexible and lean manufacturing.

2. Background

The concept of structural complexity was adopted from Shannon’s information theory [Shannon 1968]. It provides a measure of the amount of information associated with the occurrence of given states. The ‘entropy’ has its origins in thermodynamics, however, in the context of information theory, entropy is defined as a measure of the uncertainty of a random variable [Applebaum 1996].

Given a set of events and the prior probabilities of each event, where and , the entropy function of system S can be written as:

where Hs is the entropy of system S consisting of N different states, from i=1, …, N.

Pi is the probability of the system being in state i

Borrowing the notions of information theory, manufacturing complexity can be used to describe the intrinsic difficulty of a system, the information content of a system, or other information and uncertainty related management problems.

This paper adopts the definitions of complexity in manufacturing as “the amount of information required to describe the state of a system” used by many authors [Deshmukh 1993; Frizelle and Woodcock 1995; Calinescu, Efstathiou et al. 1998; Efstathiou et al. 1999; Calinescu et al. 2001; Arteta and Giachetti 2004]. The structural complexity can be interpreted as the degree of variety and uncertainty which have been fully discussed in many literature [Frizelle and Suhov 2001; Sivadasan 2004; Yu and Efstathiou 2001; Huatuco 2002].

Benjaafar proposes a similar measure for assessing the information related to the processing and product flexibility in a flexible manufacturing system [Benjaafar 1992]. Benjafaar’s work has an advantage to expand for process plan analysis due to its simplicity.

The concepts of value-adding complexity and non-value-adding complexity have been brought out by [Calinescu, Efstathiou et al. 1998] and [Huatuco 2003]. They show that the complexity could be value-adding or not, depending on the nature of the problems that gave rise to dynamic complexity.

To date, these works provide very little implications and insights to network system due to the inability of measuring complex system. This weakness significantly limits the applicability and comparability of entropy measure. In order to convert the results into meaningful suggestions for system re-engineering, extensive works should be focussed on the inter-dependency of machines and some special structures such as rework and buffer system.

3. Manufacturing system re-engineering

One of the first widely available books on re-engineering [Hammer and Champy 1993], gave a clear definition, although they misses the importance of complexity:

"Reengineering is the fundamental rethinking and radical redesign of business processes to achieve dramatic improvement in critical, contemporary measures of performance such as cost, quality, service, and speed."

A number of researchers focus on the application of mathematical models on the re-engineering problems. A table of these re-engineering problems and the related works are listed below.

Table 1 Classification of system re-engineering/optimisation problems

Descriptions / Reference
Throughput optimisation by allocating of component reliabilities / [Hillier and Boling 1966; Misra 1972; Beraha and Misra 1974]
Throughput optimisation by redundancy allocation / [Tailman et al. 1968; Misra 1973; Misra and Sharma 1974; Lee et al. 2001]
Cost-effective optimisation by component assignment or redundancy / [Hillier and Boling 1966; Dolins 1992] [Hussain and Murthy 2000]
Optimisation of complex system / [Seidel 1983; Murty and Reddy 1999]
Application and case studies / [Avadhani 2003; Liu et al. 2004]

Apart from quantitative analysis, we can take a broad and more qualitative perspective in solving the re-engineering problems, especially two popular methods that are highly adopted by industry currently, which are lean and flexible manufacturing.

Lean manufacturing derives its name from the manufacturing systems and processes of the Toyota Production System. These lean systems are very effective at producing at low cost, high quality, and short cycle times [Bunce and Gould 1996] [Feld 2000]. The key to making the transition to "Lean" is to make the processes flow seamlessly [Womack and Jones 2005]. Waste elimination is one of the most effective ways to achieve the lean transition. Practically, it can be equivalent to a famous industrial management principle, “Seven Wastes” from Japan [Grunberg 2003].

Different from lean manufacturing, flexible approach is another way to gain competitive advantage by providing buffer and redundancy to deal with varying demand and unpredictable events. The job shop is a traditional design principle to allow a great amount of flexibility for process and machine selection. However, the problems of low volume, low profit and low standardisation caused by extra cost and difficulties in management are unavoidable [Hopp and Spearman 2001].

Are these two approaches mutually exclusive? It is surprisingly that very few literature attempt to compare these two methods analytically. In some cases such as using advance transportation systems and programmable machines, they can be seen as lean and flexible approach because both can reduce wastes and provide flexibility simultaneously.

In some flexible manufacturing re-engineering cases, adding parallel machines and buffers can definitely enhance the adaptability of the system. Nonetheless, they undoubtedly increase the level of waiting, processing combination and inventory which should be avoided in lean approach.

The objective of this paper is to propose the idea of using structural complexity to assist the system re-engineering. The new perspective can provide new insights on handling the trade-off problem between flexibility and throughput. Moreover, the complexity value can be interpreted as a key indicator to distinguish lean and flexible manufacturing.

4. New complexity measures

In our previous works [Yu and Efstathiou 2001, 2003, 2005], four new structural complexity measures have been developed including network complexity, buffer complexity, routing and sequence disorder complexity. The new development extends the capability of structural complexity to more complex structure and system behavior. A brief summary of them is listed next.

1.  Network complexity extends the application of structural complexity to multi-component system by considering the inter-dependency between two components. The idea of inter-dependency can be explained by a system installed with buffer. If there are sufficient stocks at buffer storage in a two-machine series system, the status of one stage is nearly independent because its status is not affected by the other one. On the other hand, the independence will gradually decrease when fewer buffers are used. Using aggregation and reduction methods, it is possible to extend the network complexity to examine more complex network structures [Yu and Efstathiou 2001].

2.  Buffer complexity has been developed to assess the average amount of information associated with the variation of buffer. In a manufacturing system with a buffer installed, the stock level of buffer will go up and down occupying different states, which brings the problem of uncertainty and variety. Using Buzacott’s buffer model [Buzacott 1968], it is able to determine the probability of each buffer state using system characteristics including arrival, reject and repair rate.

3.  Routing complexity is similar as Benjaafar’s flexibility complexity. Contrary to flexibility complexity which focuses on the processing options, routing complexity concerns the choice between alternative routes, instead of processing options. It represents the effort required for management and monitoring production flow [Benjaafar 1998].

4.  Sequence disorder complexity measures the degree of disorder in a rework cell by considering the probability of each disorder state [Yu and Efstathiou 2005]. For most reworking processes, parts are delayed by the rework process after they fail inspection. The repaired items will appear in new positions after rework. See [Flapper 2002] and [Campbell and Mabert 1991]. Their works show that the time and material losses in cyclic production schedule would cause certain products to be unable to follow the preset sequence schedule during certain periods.

Please refer to the corresponding reference for the formulation and other details of each measure. This paper mainly presents the findings and implications obtained by these measures and the implications to system re-engineering.

5. System re-engineering using complexity

As mentioned in the abstract, four particular re-engineering ways are investigated, namely layout re-structuring, buffer size adjustment, modifying the parameters of individual station and rework cell re-configuration

5.1 Layout re-structuring

There is a large number of ways to re-structure a manufacturing network. The investigation begins with the simplest forms, serial and parallel structure. Serial configuration is the sequential relationship between a predecessor and a successor. A breakdown of any station would make the whole system halt. As a result, reliability is one of key problems in the management of a long serial production line.

Merging two similar operations is a common approach to simplify the system. By identifying each process in the system, they can then be re-categorised into function groups. Reduction or elimination should be taken to avoid unnecessary steps. The lean approach can cut down the cost of safety stocks and save resource for extra operation [Feld 2000; Yan et al. 2002]. The other important issue is the lessening of uncertainty regarding process control when the production line is shortened. This can be explained by observing the complexity change after merging stations.

Figure 1 An impact of merging stations on complexity and throughput at A=0.93

Figure 1 displays the complexity value of a multi-station system given that A=0.93. It is generally advised that merging operations in serial production line can reduce the complexity as shown in Figure 1. Such drop in complexity can be seen as one of impacts of lean manufacturing to trim down the scale of production, in particular to avoid those unnecessary steps. The usefulness of dividing operations is overshadowed by merging operations due to the extra complexity which make it unwelcome to system design. However, under some circumstances, dividing operations could be effective by utilizing the concept of division of labour. By specialising jobs into several operations, dedicated workstations can focus on their own tasks. Hence, the completion time of each task may be reduced and throughput may rise. It is recommended that the effect of division of labour should be considered at the same time when a merging operation is performed.

In many plants, machine breakdowns are one of the largest and most disruptive problems, particularly in series systems. Adding parallel stations in vulnerable stages can increase the overall reliability by providing redundancy. The operation can continue running as long as one of the parallel stations is working. Despite the advantages brought by redundancy, it induces extra cost and complexity related to the extra flexibility in the system.

To understand how effective parallel stations can contribute to system re-engineering, it is useful to examine the trade-off between complexity and throughput after adding a station in parallel. Two major models of parallel production line are compared and analysed; the first model is a redundant model. It is assumed that the system functions at its full speed when at least k of its n components function. The system will work at less than the maximum rate if fewer components function. The second model is a non-redundant system which all components perform the same function. If one breaks down, others continue working, but the production rate is reduced to the sum of the working components. This section goes as follows: redundant model, non-redundant model and comparison.

Figure 2 An impact of changing the number of parallel stations on complexity at A=0.8

By varying the number of stations, n, we compare the complexity value of non-redundant model (n out of n) and several (k out of n) redundant models with different values of k in Figure 2. It is clearly found that the non-redundant model has the highest value of complexity for all values of n. It is interesting to observe that the beginning part of the curves of redundant models joins the curve of the non-redundant model. This is caused by the idea that an n out of n redundant system is equivalent to the non-redundant system which we discussed just before. For example, 2 out of 2 and 6 out of 6 redundant models are the same as non-redundant models with 2 and 6 components respectively.

The tools of assessing complexity and throughput in serial and parallel structures have been developed in this section. It provides a direct analytical comparison on the impact of simple structural modifications such as adding and removing stations in serial and parallel configurations. In addition, it is possible to use this tool to analyze the sensitivity of availability and trade-off problem between complexity and throughput.