Grup teknolojisi imalat sistemi ve sezgisel kümelendirme yöntemi

Abstract

Grup teknolojisi (GT) sistemi grup yerleştirme ve malzeme akışım basitleştirmeye dayanan parti üretimine yeni bir yaklaşım olmuştur. Kütle imalatında hat düzenleme ile sağlanan avantajlar, parti üretiminde bu teknik ile elde edilmeye çalışılmaktadır. GT küçük parti çok çesit durumunun mevcut olduğu her endüstride kullanılan bir yöntemdir. Bu yöntem çeşitli parçaların az sayıdaki miktarlarının benzer karakteristiklere göre gruplandırılmaları şartıyla daha ekonomik olarak üretilebileceği gerçeği üzerinde durmuştur. Bu çalışma grup teknolojisinde literatürde yer alan önemli kümelendirme algoritmalarım incelemektedir. Ayrıca pek çok kümelendirme algoritmasında olmayan, parça rotalarını ve bir makinaya birden fazla defa uğramayı gözönüne alan sezgisel bir yöntem incelenmiştir. Bu yöntemin endüstri uygulamaları için oldukça kullanışlı olduğu anlaşılmıştır. Ancak her isletmenin kendine özgü şartları gözönüne alınarak en uygun yöntemin seçilmesi gerektiği unutulmamalıdır.

Group Technology (GT) is the philosophy of identify ing and gatering similar components to take advantage of their similarities in the design and manufacturing proces ses. Implementation of the GT requires the definition of a suitable measure of similarity such that the items can be grouped into families by an appropriate clustering technique. In general, the objective of GT is to trans form a job-shop scheduling problem into a number of flow- shop problems. By using GT, parts with similar attributes can be classified and coded into parts family to simplify and speed up process planning and manufacturing. The benefits of utilizing GT in the entire manufacturing pro cess has been reported to have considerable savings in parts design and drawing, production and quality control costs, setup and troughput time, raw materials, work-in- process and finished goods inventory. GT with its underlying principle of exploiting parts similarities in all stages of the production cycle, pro mises to be a major tool in answering to the problems of low volume, high variety production. GT is an organized principle that promises widespread benefits to small me dium batch manufacturers. Batch manufacturers are frequ ently facing smaller lot sizes, larger varieties of pro ducts and are thus unable to take advantage of mass pro duction techniques. The ideal situation is to be able to produce a large variety of component parts on standard machine and at the same time achieve the productivity and economy of mass production. One approach is the analysis of the material flows and machining requirements of the components constituting a company's products. Families of components ar& identi fied based on similar machine requirements. These machi nes are brought together into groupes called manufacturing cells. A method called machine-component group analysis is sufficiently accurate to indicate to the company the scope for rearranging the shop floor into independent ma nufacturing cells. The basic input data is the list of machines that each component visits, ignoring the exact visitation sequence of those machines. Using route card data, a machine-component matrix is prepared, in which the rows represent machines and the columns represent compo nents- or vice versa. If a cell entry A^.j =1, it indicates that machine "i" makes component "J", or if A-u =0, there vi is no relation between the two. So, the complete matrix is a random array of O's and 1's. The clustering algorithm rely on the basic assumption that the group of machines and components can be partitioned into matched group of machines and components. Those will be represented as clusters along the diagonal of the matrix. By using the Rank Order Clustering Algorithm, the card data are repre sented as a binary matrix. Using a positional weighting technique for the "1" entries in the matrix, the rows and columns are alternately rearranged in order of decreasing rank. The result is a diagonal ization of the 1's into se veral clusters. The production-oriented approach can be further divi ded into several categories with respect to the differen ces in clustering logic. Among the array-based methods, the ranked order clustering has been used most frequently, and among the hierarchical clustering methods, the single linkage method has been used most frequently. A hierarchi cal clustering method must compute the similarity and dis- similarilty between each pair of parts of machines in 01 - der to produce a dendogram for final judgement. Whatever it is a cluster alogorithm must assign a clustering criterion as an objective function in order to optimize system performance. The criterion could be simi larity or dissimilarity index derived from binary or nume ric data. A similarity coefficient is used to measure the degree of similarity. The larger the coefficient, the higher the degree of similarity between each pairs of parts or machines. A dissimilarity coefficient, converse ly, measures the degree of dissimilarity. The dissimila rity can also be defined as the distance between two clus ters. Finally, the results are quite varied from algorithm to algorithm and require subjective judgement of human in teraction. GT has been recognized as the key to improve product ivity, material handling, management and control of a typ ical batch-manufacturing organization. GT can also simp lify the design and the process planning of new products by taking advantage of similarities in part design and ma nufacturing characteristics. The basic idea of GT viewed solely from a manufacturing viewpoint is the decomposition of the manufacturing system into subsystem, by classifying parts into families, and machines into machines cells, ba sed on the similarity of part manufacturing characteris tics. Parts that have to undergo similar operations and require a given set of machines for these operations are grouped toqether. These machines* are in turn grouped into machine cell, thus forming a subsystem or cell. It is im portant to note that design characteristics alone cannot usually be used to determine the manufacturing subsystems, because the manufacturing characteristics of seemingly si- vn i milar parts may be entireely different (and vice versa). Part classification, however, from a design point of view, can be useful in simplifying the design process of new parts. To find a completely unencroached partition, wherein each part of a part family remains confined to one machine cell is in fact an illusion in a typical industrial appli cations. The presence of alternate process plans, dupli cation of machines, and subcontracting may not always eli minate the problem of inter - cell transfers. Further, the problem becomes more aggravated in the case of make-to- order parts. Most of the suggested algorithms in the literature ar& either not amenable to problems of a large size or are computationaly prohibitive in typical industrial applica tions. Furthermore, they do not address the following is sues. 1) The sequence of operations. A typical matrix foi - mulation clustering approach attempts to confine the operations of a part to a corresponding cell of machines, regardless of the sequence of opera tions. However, the sequence of operations in a part production routing has an important effect on material handling costs and times. It is evident that an intermediate operation of a part performed in an external cell will necessitate two inter - cell, is just as bad as having a single outside operation, since the material handling effort in volved is the same. 2) Non-consecutive operations on the same machine. It is not uncommon that the same machine is used more than once in a part routing: and if such a part has to visit foreign cells, the implications on material handling are significant. Matrix for mulation approches cannot address this situation because each element of the incidence matrix can hold only one entry. The proposed heuristic addresses the above issues, and recognizes that a perfect decomposition is generally not possible. It attempts to determine machine cells bet ween which the inter-cell traffic is minimum. The procedu re is twofold. The first step is a bottom-up aggregation which formulates a basic assignment of algorithm, provided the size of the union remains within prescribed limits. The second step is a refinement procedure in which the ba sic assignement is improved if possible. In this step, it is ascertained that each cell is assigned to that cell for which it is it is the most significant. In this thesis a twofold heuristic procedure is pro posed to reach a good, if not optimal solution. At the be- vm ginning, each machine is placed in a separate cell. At each step of the minimization procedure, the Normalized Traffic for each feasible aggregation is calculated. Feasible aggregations are those which, if allowed, will consist of a number of machines not to exceed the cell- size limit, N. The two cells between which the Normalized Traffic is maximum are aggregated into a single cell. Eve ry aggregation is accompanied by a reduction in a number of cells by unity. Subsequently, the total inter-cell traffic in the system will decrease by the value of traff ic between the two component cells of the aggregate. The traffic between cells is now revised by the following ru les : 1) The traffic between two unaffected cells remains the same. 2) The traffic between an unaffected cell and the new aggregate is the summation of the traffic between the former and the components of the aggregate. The procedure is continued until it is either not po ssible to perform any feasible aggregation, or the traffic between each of the existing cells is zero. At the beginning of the refinement procedure the sys tem is partitioned into machine-cells. Then the total intra-cell traffic in the system is tried to be partioned basically. In other words, the total intra-cell traffic is converted into mintra-cell traf fie. Reflecting back to the basic assignement procedure, machines are grouped signifi cantly, but once a machine is assigned to a particular cell, it cannot be reassigned to a newly aggregated cell. even if it is more suitable for the aggregate. Thus, in the refinement procedure, at each step of the algorithm one machine is considered as a separate entity, and its traffic with each of the existing cells is evaluated. This machine is assigned to the cell with which its interaction is the most significant. The process is repeated for each machine. Usually machines are reassigned to the same cells as in the basic assignment, but it is important to note that reassignments are possible. Reassignements also impact on the natural size of the cells. In this study, a simple but effective method of grou ping machines in order to minimize inter-cell material mo vement is presented and tasted with a variety of inputs of varying degree of encoachement. Respecting the sequence of operations, inter-cell -traffic is attempted to be mini mized. It is observed that in a low degree of encroach ment, it is possible to find optimum results. However, if the input is highly encroached, the quality of the basic assignment is sensitive to the user-friend limit on machi nes per cell, and to whether or not it is relaxed during the algorithm. The improvement algorithm is capable of refining the basic assignment and in most cases however, ix it may result in just a few Targe cells. Thanks to the high speed of the algorithm, a few enumerated runs can be made inexpensively. The order of complexity is almost in sensitive to the number of parts "n". The proposed system can also take into account multiple non-consecutive opera tions on the same machine, a case which has not been addr essed by existing clustering methods.