Production planning, MRP, production control, routing and scheduling - of these scheduling is one of the ugliest secrets on the supply chain. Its role in the management of capital assets and inventory is fundamental, yet traditionally it has been an orphan, certainly in terms of plant management. Vast sums are sunk into design and manufacturing engineering, yet scheduling, which after all is the economic use of these resources, attracts comparatively little investment and is frequently delegated to junior managers with even less bargaining power in the organization.
Fine balance
Scheduling appears again and again on the supply chain. The co-ordination of material supply, transport resources and warehousing all obviously depend on scheduling activity. In keeping with the dynamics of the automotive supply chain, however, the pivotal resource is that of the assembly plant and perhaps, as a consequence, it offers the greatest difficulties in term of scheduling. The location of some of the most significant capital assets on the supply chain, it is also represents the most important 'choke-point' in terms of the ability to respond to customer demand. Investment in paint shops or welding machinery can run into tens of millions of dollars so it is vital that it is utilized to its maximum. This is equally the case with the large workforces employed at assembly plants. The traditional response has been to increase the length of production runs to assure the highest utilization of capacity. The problem with this is, what is being produced may not be what the customer wants. The result is a build-up of finished product. This illustrates that there is a tension between two imperatives - the need to achieve consistently high utilization and the need to respond to often varying market demand. A successful schedule is one that finds the optimum between these two drivers.
The problem
The demand for vehicles is highly variable. Customers are offered a whole series of specifications in terms of colour, engine size, body type and so on. To satisfy these, the line uses a number of different activities, such as use of different paint shops, welding locations and the addition of different types of components. To satisfy demand the vehicle must pass through these locations and actions. With all of these variations, the number of options is huge and such variability is extremely difficult to manage. Traditionally, any scheduler has two basic options: avoid the problem by arranging facilities to cope with a smaller more predictable range of product; and attempt to create a mathematical model capable of rationalizing the options available.
Usually the most wise path is the former. Creating a model which truly reflects all of the possibilities that face the schedule is impossible. Yet PA Consulting claim they have been able to do it.
Nissan's production line, Sunderland, UK plant
The solution
The approach used by PA Consulting Group at their project at Nissan Sunderland was to include 'constraint logic programming'. This is a technique that is used elsewhere in the automotive sector, including by other VMs for scheduling, but PA point out that Nissan had a uniquely complex problem due to the movements between the lines (see Figure above). Their approach requires the following input data to be available:
- Volume plan - How many and of what type of model should be built in a given period of time
- Calendar- working period
- Order set - the order in which individual model should be made
- Production rules - the limitations on the ability of the line to respond to changes in the type of each model
- Line map - the layout of the line and the location of each assembly action.
Next a 'route' is traced through the plant for each car that has been ordered. This is done by taking the volume plan and the calendar and reconciling it with the available production slots on the various areas on the line, the latter being worked out by combining the order sets and the production rules. With so many variables working together this would normally be too difficult a process to calculate. However, the key to PA's solution is a series of algorithms that can cope with such complexity.
The consultancy is reluctant to disclose the details of these algorithms but PA's Scheduling systems consultant, Iain Maclean commented that, "The realisation that we could do this was fundamental to the solution".
The essence of 'constraint logic programming' is its ability to cope with, not only a high number of variables, but a series of different situations, each one having a number of variables, so that in this case the action on the production line is continuously changing. The new system translates a multi-dimensional problem into a single formula requiring one set of inputs and producing one set of outputs. At its simplest, the production resources required to build each model passing down the line is reduced to a value reflecting the amount of work that needs to be done. "So you look down the list of vehicles that will fit to what is available on the line", commented Maclean. An aspect of the scheduling solution is its ability to deal with the multi-dimensional relationships of different orders - in short to find places on the production line that will not conflict with other orders. "As you make a move you are working out the effects of that move before making the next one" . The consequent values and algorithms have been loaded into an IT system using ILOG 'Solver' software as the engine driving the calculations.
Nissan - a masterpiece of complexity
Nissan's production facility at Sunderland, in the north-east of England is probably the most efficient assembly operation in Europe. It consistently tops the Economist Intelligence Unit's table of output per man. Moreover, this productivity level is achieved without compromising the quality of the product, which again rates highly in terms of reliability and build quality.
However, Nissan have had to face problems in sustaining this. Modest market share has meant that the volumes do not exist to sustain dedicated plants or assembly lines for each model. By the late 1990s Nissan were confronted by the need to produce three models in one plant. The option existed to create another line at Sunderland, but this would lower the plants productivity levels. So Nissan designed lines that were capable of producing multiple models. The result was a complex production facility, characterized by movement from one line to another.
The facility may have been an engineering tour de force, but Nissan quickly realized it was a scheduling nightmare. While many plants have a very large variation in the specifications of models produced, few, if any, the problem of handling three different models on two lines, which interact with each other. Nissan realized that it would have to make enormous developments in its approach to scheduling.
The approach taken was brave, some might say foolhardy, but Nissan had little option having already built the lines. What the company sought to create was a mathematical approach which 'modeled' the assembly line by collating its limitations, thus leaving the option of what could be done. This is what is called 'constraints management'. To help them do this Nissan employed PA Consulting Group who have specialised in the creation of the analytical tools required.
Proof in the system
Implicitly they claim to have created a new analytical tool that is capable of modeling complexity to a level of accuracy that changes the potential of such tools in solving scheduling type problems. Even if PA did disclose the details of the algorithms it would be difficult to assess their voracity other than by looking at how effectively the plants capacity is managed. Maclean, however, offers evidence of the effectiveness of the system. In early trials the new scheduling system highlighted that the volume plan was not accurate. This came as a major surprise to many at Nissan, but it helped to explain why the douki sesan measure (see below) was so low. Consequently, the joint PA and Nissan project team developed a new technique for developing routes which worked backwards from the order requirement. As the routes indicated when and which models would pass through each point in the plant, it was straightforward for the volume plan to be gathered from this information - so the volume plan changed from being an input to an output of the scheduling system.
The result of this richer information picture is a more predictable production environment. "Nissan has a measure called douki sesan, an off-line point which measures how many vehicles go through that point within plus or minus one hour of when they should have gone through. If they got to three per cent of that on their old system they were happy. The new system gave them 85 percent in the first week of use and now averages 93 percent. It is much more accurate about when [vehicles come off the line]", commented Maclean. The result of this greater predictability is a reduction in uncertainty, which in turn has lead to lower inventory levels of inbound materials by about half.
A failing of mathematical approaches to these sorts of problems in the past has been their inability to comprehend the level of variability in the production environment. Once such systems have been up and running a series of 'bugs' emerge as the system is unable to cope with unforeseen circumstances. Maclean admits that did happen at Nissan. For example, he cited the fact that initially the need to cope with varying paint quality by limiting the size of the batches through the paint shop was appreciated, but he states that such problems were sorted out within three weeks. It has also proved to be highly reliable, "producing the Nissan car plant schedule every week since August 1999".
Weaknesses
Naturally PA are proud of their system and regard it as having few weaknesses, however no solution is perfect and this is no exception. One possible failing of constraints logic programming that Maclean cites is a lack of diagnostic capability. "It is", he says, "unable to generate recommendations. When a scheduling solution is not found…the system cannot tell you why it cannot find a solution, it can only tell you that it has not found a solution. As a consequence, applications involving constraints logic programming require user familiarization".
Also, such a sophisticated solution does require access to strong human resources. While PA insist that it does not need extraordinary people to work the system it does " require expertise with the system, you cannot have just anyone use it". Crucially it also needs a manager who both understands the operation and can comprehend the mathematical concepts, what Maclean describes as a 'test pilot…a guy who is able to write the operational manual". These have to be regarded as important questions as the scheduling system is so important to the plant that any company has to able to understand how it works and to be able to take possession of it.
Supply chain context
An assembly plant is only as effective as the supply chain in which it exists. Not so much a failing as a limitation to the project is that it is concerned with optimizing the internal plant operations and does not address other areas on the supply chain which have a substantial impact on the effectiveness of plant operations, notably the capture of orders at the dealerships and the supply of components from suppliers. It is unknown whether the greater flexibility of the plant results in a more difficult forecasting environment for suppliers, although the better quality of information has reduced inventory at the plant. MacLean does make an interesting observation, however, about how such systems relate to other parts of the supply chain. "You can't build one single system to do that, what you have to do is identify clear borders between one set of processes and another, and optimize each of those….otherwise you can produce a sub-optimal overall answer. You have to look at the overall problem and work out which parts you don't want to optimize. If you try to optimize them all you will produce a sub-optimal answer for the whole plant." This is an important insight as it raises the further question of how to ascertain the effectiveness of the plant in the context of the supply chain.
PA insists that the Nissan project is fairly unusual due to the sheer complexity of the plant. However this project is still important. Vehicle manufacturers are increasingly eager to create more responsive production capabilities yet are faced with demand for more variation in product lines. Order-to-Delivery initiatives and the need to get a grip on the management of finished product inventory demand a far more rigorous approach and something that appears to represent a significant leap forward in scheduling capability demands a closer look.
Wider applications of scheduling capability
Whilst this project was uniquely complex for the automotive sector, PA Consulting Group has also applied its knowledge of constraints planning in the transport sector. An early project of this type was exercised on a resources management system for British Rail and the company has recently concluded a project for the Port of Felixstowe. The latter was designed to improve the utilization of cranes and container handling equipment. This was part of a wider system to improve the generation and distribution of data through superior data capture hardware and software. This illustrates the role that such technology plays in scheduling problems.
