Production Control of Fibres Manufacturing Processes
December 9th, 1974
Contribution to the Institution of Mechanical Engineers Discussion Meeting on “Use of Computers for Control of Production”
S F Bush
The major problem treated is the control of product uniformity in a multi-stream sequence of manufacturing processes with material splits and blending operations interposed between the stages. While fibre-making is in some sense an extreme case, it is, in a number of instances, an important exemplar of component manufacturing and assembly processes. Regarding the material from each of the process streams (which can number several thousands) as components for the finished product (woven or knitted fabric in this case) means that tolerance on critical properties across and within streams has to be of the order of 1% to maintain finished product rejection rates below for example 5%.
To obtain this performance, computer-based systems have been designed to implement a combination of five potential forms of production planning and control: diagnosis of machine and raw material faults, filtering of product stream differences, down-stream compensation of product faults, product mixing and blending, and process segmentation (division of each stage into blocks of equipment with similar measured performance).
The major problem centres on the not uncommon situation in which quality can only be finally assessed by customer experience of a product assembled from thousands of components (in this case fibre filaments). The overall system aimed for is thus one in which feedback of information from the market is an integral part of control. However, since the time delay for such a control loop is typically of the order of half a year, such a feedback can only be successfully used to monitor changes with a time-scale several times this and in practice this reduces to (a) monitoring market trends, and (b) surveillance and adaption of the business control system itself.
Within this outermost loop, there are several product quality measures available with time-scales in the range from a few hours to several days. These time-scales are still too long to permit feedback control of the process. Thus a combination is used of the five forms of essentially predictive control listed in the second paragraph. Of these, diagnosis of process and machine faults appears to be the most promising for development on a number of counts: (i) it is perhaps the commonest managerial day-to-day action and acts directly on product quality and machine costs, (ii) suitably organised data-banks are potential repositories of operational experience and machine performance, and (iii) learning to recognise pattern in data promises to be a fruitful extension of current statistical analyses of production.
In a multistage process, faults entering a product stream tend to be transmitted from stage to stage unless specifically compensated for. In general, faults and variations will have time dependent patterns associated with the characteristics of the faults themselves or the process. Such patterns are the basis of much human judgement in the situations described by means of computer-based systems of analysis. The overall system will automatically take advantage of improved information resulting from, for example, measurement innovations, or the elimination of stages.