In the Como region north of Milan, just a short distance from the Swiss border, an expert system called Setex assists La Braghenti SpA, one of the area’s oldest manufacturers of silk, in optimising its production of natural fibre fabrics. Supplying largely to the top fashion houses, Braghenti has an average of 6,000 fabric orders […]
In the Como region north of Milan, just a short distance from the Swiss border, an expert system called Setex assists La Braghenti SpA, one of the area’s oldest manufacturers of silk, in optimising its production of natural fibre fabrics. Supplying largely to the top fashion houses, Braghenti has an average of 6,000 fabric orders in progress at any given moment, of which 65% is for export. The profit margins and time to market are miniscule. The prompt arrival of the fabric is more important than ever. Years ago, a supplier produced a high-quality fabric and another produced fabric of lower quality. But now the average quality is acceptible, so the game becomes one of few lire and little time, says Giampiero Cappelletti, production manager at Braghenti. In its factory containing $3m worth of materials, Braghenti produces up to 100 different families of fabric every season, which are classified according to the their assembly on the weaving machines and the number of threads per centimetre.
Wool and linen
The [fabric] family becomes a strong point of reference for the planning process, Cappelletti says. The weaving machines at Braghenti use pre-dyed thread, which means that the company must wait for a customer’s order before weaving the fabric, Cappelletti says. This creates problems for managing the production variables. Every six months, we have to manage 400 variables, of colour, of the thread’s material. The number of possible combinations is unimaginable, he says. The procedures to acquire the raw materials are the same, but to produce and sell wool and linen are completely different things. At the beginning, we didn’t how much they would be able to solve our problems, Cappelletti said. It’s no wonder that they wondered, given the intricacies of such a scheduling system for a fabric manufacturer: the difficulty of executing a portfolio of numerous requests, which don’t comprise a unit, but are still globally consistent; the high number of unanticipated orders and urgent requests; the absolute respect for delivery dates to a market that typically burns up a product in a short period of time; the fragmentation of the production into numerous economic lots, making it difficult to distribute costs and set-up time over an optimal number of product units; and the complexity of managing a workload that comprises hundreds of articles and dozens of colour variations. As it was, it took between three and four years to perfect the system, which can be described conceptually as a job-shop-scheduling system, over which are imposed the required delivery deadlines and minimising of set-up time, says Guido Torriani, a now-retired Bull Italia SpA engineer who was responsible for development of the Braghenti system.
By Marsha Johnston
Torriani was responsible for Bull’s artificial intelligence research effort that began in the early 1980s. Around 1988-89, we gave up on this idea to emulate the human mind and we chose a certain class of problem – dynamic planning. We concentrated our efforts in this area because it was a potentially vast market and the mathematics for such a problem had been well established. At a certain moment, we realised that, although we thought the technology was resolutive, there were two phenomena that it could not master, he recounts. The first was the quality of the solution. Even though the system emulates a person, the person is always going to search for the most acceptable solution, with that acceptability not easy to define. We could emulate the process of that search, but we couldn’t approach the process of choosing one solution among thousands. The second phenomenon was tied to the know-how in the sector, he continues. We realised that there is know-how that the system cannot acquire. Such know-how is formed through personal contacts, presupposing a daily sensitivity to the market that the machine doesn’t have. Braghenti’s Cappelletti says, Our boss, for example, sometimes succeeds in forecasting what the market is going to demand. We can’t precisely define or measure what he say
s or where he gets the idea. It comes from visiting and talking to customers. So, we forgot the idea of entrusting everything to the computer, and instead we created a dialogue between the operator and the computer to search for a solution, Torriani relates. The Setex system receives real-time control data from sensors on the weaving machines, measuring the thread already used and the amount that is left to weave.
The system also contains the most important criteria for optimising Braghenti’s operations, which regard the importance of the customer from the point of view of delivery time and product quality required. As a result, says Cappelletti, the system tells me in real-time that, if I go in a certain direction, these will be the ramifications on the company’s operations or on orders received. Says Torriani, The machine is most valuable for putting together all of the variables, something of which the operator isn’t capable. If the operator changes one of the system variables, such as reducing the amount of delay acceptable for one order, the system recomputes all of the schedule to comply, suggesting, for example, the addition of another weaver. The system also helps Braghenti to change its mix of fabric families in a way that involves the fewest possible number of changes of weaving machines from one family to another, which is a major effort, Cappelletti explains. It suggests a solution that is not always the best, but isn’t the worst, either, Cappelletti says, adding that Setex now saves the company approximately $200,000 per year in increased productivity and in reduced late delivery charges. Besides, without Setex it would be impossible to have all of the real-time information we need, he says.