Jump to content

- - - - -

Suppy Chain and Artificial Intelligence

  • Please log in to reply
2 replies to this topic

#1 TheBigDog


    Doing the Impossible

  • Members
  • PipPipPipPipPipPipPip
  • 4095 posts

Posted 01 April 2007 - 07:24 PM

Part 1

For the past several years I have been working in an IT department as an expert on shop floor operations. During that time I have learned a great deal about Just In Time (JIT) manufacturing. For the past year-and-a-half I have been working on a project entitled Advanced Planning and Scheduling (APS). This is a rather generic term for software designed to organize your supply chain, helping you reduce costs on several fronts by smoothing your operation. This post is an examination of how this software utilizes advanced algorithms to try and run a supply chain, and where this system falls down when compared to the computational power of the human mind.

First, some basics about supply chain planning...

Generally speaking there are two schools of thought about supply chain planning: Mass Production and Lean.

Mass production is characterized by reducing cost through high speed production and long runs of product. In this environment you use forecasting to predict your sales and have products on the store shelves before they are ordered by a consumer, or what is called a "push" system. The problem is the accuracy of forecasting you risk pushing more or less than the market demands. If you forecast too high you end up with too much inventory, and you have manufactured (invested time, energy and money) in product that sits on shelves and represents a loss. If you forecast too low you end up leaving consumers wanting, and losing sales to competition when you cannot deliver product under your customer's normal service expectation. Imagine you are making bar soap. It takes 2 days to change your machines from one product to another, so you only change setups every month. Your product sells out in two weeks, but your production schedule doesn't allow you to restock for two full weeks. As much as people love your product, they are not going to skip two weeks of having soap to wait for your product to return to the shelves. The manufacturing companies that use long run mass production cause a dependency in their supply chains. When a factory is ordering raw materials in month long batches, the suppliers set up their operations to be optimized for the same long lead times.

This makes it impossible (or very expensive) to change production plans because even if the soap company wanted to, their raw material suppliers are not ready to change schedules on a dime.

In a Lean Manufacturing environment you want to have no inventory. You never want to manufacture a product until a customer has ordered it. Instead of utilizing forecast, lean uses what is called a "pull" system. In a pull system you still do mass production, but instead of manufacturing the identical products for long periods of time, the company focuses on reducing the setup and change over times so they can make very short productions runs with great efficiency. They strive for what is called "one piece flow" where every item manufactured can be unique from the previous without interrupting the flow of material. Taiichi Ohno was a Vice President of Manufacturing at Toyota who, while not inventing the techniques used or the theory, is the hard-assed SOB who proved it could work and turned the manufacturing world on its head. In lean the whole supply chain is geared toward smaller and smaller lot sizes, with materials arriving just in time for use and products getting into the consumer's hands with as few steps as possible.

Lean manufacturing is very agile, but you need outstanding setup and changeover performance, and suppliers who gear their production to being just as lean to get all the benefits.

As you can see, the two methods are extremely different in application. The fact is that most companies are some place in the middle. Modern companies are forced to struggle with finding a clear identity for themselves and their supply chains, and often have a mishmash of various systems and processes to try and get material out the door.

Those companies who do this the best are the ones that tend to rise to the top. An example is Wal-Mart. They are the biggest retail company in the world because they have the lowest cost supply chain in the world. They do this by having an extremely efficient replenishment system. When a Universal Product Code (UPC) is scanned on a product at a Wal-Mart store it creates a purchase order (PO) for that product to be replaced on the shelf, or adds the quantity to the existing unreleased purchase order. At a given interval (typically daily) the purchase orders are released to the supply warehouses. This is where trucks are filled with all the things needed to resupply a store based upon that periods sale. At the same time that the PO is going to the warehouse, the merchandise supplier is given a PO resupply the warehouse. Wal-Mart manages the supplier accounts very closely and keeps strict scorecards on resupply performance. When you buy a product at Wal-Mart you can literally trigger the manufacture of another one of those products within hours. Wal-Mart does use forecasting to determine what safety stock levels to keep at the stores and at the warehouses. The safety stock predicts sales and keeps enough inventory on had to insure that the supply chain has enough time to keep the shelves from being empty. So if it is twenty four hours of PO batching between releases, twelve hours of unload and stock time when a truck arrives, twelve hours transit time from the warehouse to the store, four hours picking time to fill a truck, then the Wal-Mart needs a minimum of 52 hours of product on the shelves to prevent them from getting bare. They will add a buffer to this through standard deviation of sales, seasonal variation, and shelf life to come up with a target safety stock. A similar process is followed to get a safety stock level for the warehouse. The vendor will keep a safety stock as well if they are incapable of meeting the short orders of varying sizes to meet the delivery demands of Wal-Mart. A lean company is geared to it and can keep low inventory levels for themselves where a mass production company will hold bundles of inventory and make small shipments from that inventory to meet the Wal-Mart demands. .

But enough about Wal-Mart, this is about manufacturing supply chain and artificial intelligence...

In the next post I will begin to explore how planning systems work, and how they try and think like people to make a smooth process.

  • GAHD likes this

#2 TheBigDog


    Doing the Impossible

  • Members
  • PipPipPipPipPipPipPip
  • 4095 posts

Posted 01 April 2007 - 07:24 PM


Supply Time = The time it takes from when you place an order until it arrives at your receiving dock.

Receiving Time = The time required to unload material, inspect it and receiving it.

Safety Time = The amount of buffer built into Supply time to account for unpredictable circumstances.

Setup Time = The about of time it takes to clean and prepare a machine to run a given job. Setup time may vary depending upon what sequence of products you are running. For instance going from white to yellow to orange to red to brown to black on a painting machine takes less time than going the other direction.

Downtime = The amount of time that a machine is idle for reasons other than setup (which is part of the cycle time). There is both planned and unplanned downtime. Planned downtime includes Preventive Maintenance, no orders, plant calendar days off, and any other reason you might plan. Unplanned downtime is primarily breakdowns. Running scrap does not count as downtime, and is a common mistake made when calculating production efficiency.

Cycle Time = The time needed to process a single lot of product. This includes all manufacturing steps, such as a car that may stop at 100 assembly stations, the cycle time is the time from beginning to end of the whole process.

Throughput Rate or Production Rate = the number of units manufactured in a given period of time.

Scrap = Anything consumed into a manufacturing process that is not part of the final product. This is not just material, but time and energy as well.

Bottle Neck = the process in any multi step manufacturing that has the slowest throughput rate. If there are 100 steps to manufacture a car, you cannot make them any faster than the slowest step (you would be surprised how difficult a concept that can be).

Takt Time = The process of getting all manufacturing steps to happen at the same rate. Takt time can refer to both that established rate and the process of achieving it.

Muda = Japanese term for waste adopted by lean manufacturing to identify unwanted processes.

Safety Stock = the about of something you wish to keep in inventory at all times as a safeguard against running out before you can get resupplied.

Reorder Point = a preset quantity that triggers ordering more supply

Forecast = A prediction of future usage used to plan operations beyond your order horizon. This is most important when the lead time you promise is shorter than the lead time for you to receive materials to make product.

Vendor Managed Inventory (VMI) = When a supplier takes over the whole supply process for a customer to lower the operating cost for the customer and allow the supplier to smooth their own operations.

Order Horizon = the average lead time for orders you take to be delivered. *this is my own term for understanding how planning systems make the transition from planning reality to planning forecast.

Minimum Order Quantity = The minimum sized order that a company will accept for a product. (There are several order types, and this can be defined at each level)

Order Increment = The step process of increasing order sizes. An order increment for eggs might be a dozen. For wholesale gasoline it might be the quantity of a tank truck.

Maximum Order Quantity = the largest order that a company will commit to at any order level. You might think this is counter intuitive, but you need to keep the order sizes small, especially when you have a diverse product mix and promises to keep to multiple customers.

Forecast Time Fence = the time on the horizon where you exclude forecast from being part of what you are going to assemble. Manufacturing to forecast is dangerous because you are not making products that have been ordered, you are making products in anticipation of getting future orders. For some product types you never want to manufacture to forecast, and in those cases you need to set a time fence so that forecast demand will not be part of your released order plan.

Order Grouping = When you get an order for a product, this represents the amount of time into the future from that order due date that your planning system will search for other orders of the same product to be included on the same work order released to production. If you have a product that you want to manufacture once a week you might keep an order grouping of seven days, so that when you make it you will manufacture seven days of material. It is done to reduce the number of setups. It can be done at multiple levels within your supply chain.

Sales Order (SO) = an order placed by your customer for your product. This represents demand for your supply chain. Sales orders can have multiple line items, each with a different due date and item. The whole sales order is paid for under a single purchase order agreement with the customer, but each line can be delivered separately.

Sales Order Line (SO Line)= One item, one quantity, one due date, one delivery location. This is the base part of a sales order that makes the detail of what you are shipping to a customer. Planning systems consider each SOLine to be a unique demand to fill. * All order types can have multiple lines, but they may

Purchase Order (PO) = An order representing what you are getting from a supplier that you are purchasing, and there is a . This represents supply in your supply chain.

Distribution Order (DO) = An order tracking the transfer of materials between warehouses of the same company. When Ford assembles car bodies in Tennessee and ships them to Kentucky for final assembly, a distribution order is used to make this happen. There is no change in value of the product and it remains in the possession of the same company.

Work Order (WO) = An order to manufacture a certain quantity of an item with a due date. Work orders are what are scheduled to regulate and measure the production done on manufacturing machines. They are also used to calculate labor and material costs associated with production. You may also see these referred to as Manufacturing Orders and Production Orders.

Pick List = A list of materials needed for a process. This can be a list of things to bring to a machine for assembly, or a list of products to load onto a truck for delivery, or a list to pick and package for a customer order. Pick lists are the most basic work instructions for having people get material to the right place at the right time.

Routing = The series of manufacturing steps to finish a final assembly. Routings are used to calculate the labor hours.

Operation = The sequence number within a routing for each assembly point that must be completed and become real inventory.

Phantom = A product that is manufactured but is never in inventory. An example of this would be a part of a circuit board that is assembled from its own parts at the same time that it is assembled to the board. The BOM for the circuit board calls for that part, but you never stock that part, only the things that make it.

Yield = How much of an operation ends up as good material. If you have a machine that has a normal 3% scrap rate then you might set your yield at 97% so you plan to manufacture enough so that get a fair prediction of how much you really need to schedule to fill an order with good product. Yield is also found on a product structure to indicate how much of the material used ends up in the final product. You may have a product that boils during manufacture. Even though you use 100 gallons of alcohol in the product structure, only 50 gallons are in the final product when it is completed due to evaporation.

Bill of Material (BOM) = The raw materials required to manufacture or assemble a product. These are broken down for each subassembly, and each routing step is indicated. A lead time field can be employed to show relative to the start of assembly when each material is needed. You may have an operation that takes several days, and some materials are required on day one while others are required on day three. You don't want to hold up production because the day three products are not on hand day one.

#3 Buffy


    Resident Slayer

  • Administrators
  • 8946 posts

Posted 01 April 2007 - 08:09 PM

I love this topic. Used to do some consulting on it.

A key issue is the relative strengths of the suppliers and retailers. Wal-mart is an extreme example, and is especially bizarre when it comes to small manufacturers. You'll actually notice that if anything, Wal-mart has over the years deemphasized products where they have not become the primary sales channel (e.g. consumer electronics and major household appliances, where places like Best Buy, Sears, etc hold major stakes and the manufacturers have a little bit of leverage).

I've always argued that except in cases where there is very high verticalization of the supply chain--that is, that the suppliers to a manufacturer all pretty much work for that one manufacturer--that "pure lean" is not even possible: they may be separate companies, but they tied at the hip to the manufacturer and have to dance to their tune, so they do.

One thing that is often ignored in determining the "benefits" of lean, is that the costs of inventing and building the infrastructure to support it is often kept off the books, so its benefits are often over-rated. The costs of maintaining inventory are usually nowhere near as high as *both* the cost of actual switchovers *and* the cost of inventing and implementing the technology that keeps the cost of the switchovers down.

Keep goin' Big!

Fully weighted costs,