Intellection: Run your factory like an airline? – Software Review

Intellection: Run your factory like an airline? – Software Review – the Rhythm software package from Intellection is a real-time master production scheduler and optimizer

In fact, Miraglia has found his time-management solution — along with 14 other customers such as Black & Decker, Mary Kay Cosmetics, Motorola and Caterpillar. It’s a product called Rhythm, from Intellection of Dallas.[1] Rhythm, which runs on UNIX workstations, is a realtime master production scheduler. It lets a user change schedules, as do many tools. But unlike most of its competition it’s also an interactive simulation tool; it works fast enough to process a huge volume of data and perform what-ifs, reallocate resources and display the impact right away. Users can interactively assess the impact of a change in a schedule, the relaxation of a constraint or a change in supply conditions.

Right now, a worker using Rhythm can take 10 minutes to rear-range a schedule to offset the impact of a sudden breakdown on line 33. In the future, a salesperson may use it (remotely) to decide whether to accept an order at a 10-percent premium for delivery next Tuesday.

For schedule building and optimization, Rhythm uses its own patent-pending constraint-anchored optimization algorithms. [2] Developed by founder Sanjiv Sidhu and his team, they can optimize efficiently and quickly despite multiple and interdependent bottlenecks. In fact, the more complex the situation, the better Rhythm performs relative to traditional approaches.

Basically, Rhythm’s approach speeds up the optimization process by careful analysis and management of the constraints, which limits the search space of possible solutions regardless of the algorithm used to test the minima. The system first looks at an unconstrained schedule and assumes infinite resources, and then compares that to reality — the actual inputs and processing facilities or machines available.

Then it lets the user ease the constraints (for example, ordering supplies early, subcontracting work, opening up extra lines or letting delivery dates slip) and develops a schedule working from the remaining constraints out — backwards from due dates as well as forward from the present or the earli est possible availability of resources. This limits its search space of solutions to ones that are actually possible. By contrast, most systems work only forwards or backwards, and waste time considering impossible solutions.

Don’t constrain me

What are the factors Rhythm must Juggle? They include:

* set-up costs and time (How long does it take to put a machine or line into use and turn it off again? How long does it take to switch from producing one product to another? How much material, such as paint, is wasted in the process?)

* increases in cycle time as machines are used to capacity

* substitutes for resources available at any given time

* varying qualities and characteristics of raw materials and components

* labor availability and costs, including overtime and night premiums

* unscheduled (or scheduled) downtime or inventory shortages

* delivery requirements

All these factors define and constrain the range of possible schedules, and the trade-offs for each. Meanwhile, new orders come in, and old ones are modified or canceled. Rhythm has powerful facilities for representing those constraints and bottlenecks, and for calculating the priorities (as defined by users) of alternative scenarios.

Short-term and long-term

There are two basic user modules, Rhythm Master Production Planner & Scheduler (MPPS) and Rhythm Dynamic Scheduler (DS). MPPS is for planning long-term schedules; DS is for short-term rescheduling and modifications to a larger-scale plan. In long-term scheduling, the system determines the global constraints and generates the overall best schedule for a period of weeks or months — interactively with a user who can decide which constraints to relax. The system looks at overall capacity and materials needs, predicted orders and other information.

For short-term scheduling, the user checks out and downloads the necessary subset of real-time data from the existing production systems to build a new schedule — or do queries of work-in-process, maintenance scheduling, troubleshooting or other tasks. (From this point forward, a monitoring program captures all changes to the database so that integrity can be maintained.) Just as with a spreadsheet, the user can do queries, change values, reassign tasks and so forth, using the tool to see the potential impact.

When he commits any changes back to the production system, Rhythm takes the file of changes since the checkout time and runs them again against the new version of the database. Since the system operates in (real and virtual) memory, it uses the existing databases of the manufacturing operation; it’s a modeling tool or application rather than a database itself, with normal locking facilities. Aside from simultaneous users, the only major changes are likely to come from new orders, which tend to come in batches from a separate department anyway; at this point at least, customers generally aren’t online to manufacturing facilities

Fast machines, slow humans

For both MPPS and DS, the calculations are relatively swift, not just in terms of the user’s interaction, but in terms of elapsed time for producing a schedule. Long-term schedules that used to take Eli Lilly’s Cardiac Pacemaker unit 10 days to produce now take 5 days with Rhythm. Even with swift calculations, the process requires user time to consider alternative schedules, decide which constraints to relax, talk to marketing, and so forth.

In theory Rhythm plus an expert system could do all of this, but we aren’t there yet. Besides, users don’t like leaving it all to a machine; they want to propose schedules and see the impact, open a line and see what happens.

Customer comments

Apart from all the theory, users do indeed like working with Rhythm. For example, the users at Caterpillar’s gear pump division like Rhythm so much, says flow management coordinator Jim Lock, that they won’t give up their old, tested, original version of the software to install Intellection’s newest release: “They said, ‘Leave us alone. We like it fine the way it is.’ Instead, we’re putting the newest version into several other plants.”

Moreover, the system can be useful right from the start. Earl Mott, director of computer-integrated manufacturing technology for Black & Decker, recalls that during training, with real data in a test environment, planners would keep discovering problems and calling the factory to change things. The result was smoother operations before the system was even installed.

Mott’s experience is probably typical, although Black & Decker is one of the leaders in US manufacturing and helped write the spec for the current factory management standard, MRP II — which is now outdated. “Our current system is weekly-bucketed,” Mott says. “Anything happening within a week is basically invisible. Suppose we get a call from Home Depot for a thousand drills. After a day or two, all the factory can say is ‘maybe.’ But we have no idea of the impact on other orders. But Intellection gives us an extremely good view. We can call back to marketing and say, ‘We can give them to you on Friday.’ Or we can do it on Wednesday if they insist, and then we tell them what orders we have to delay to do that. Our big goal now is service level — and getting the business: A substantial proportion of our orders are ship-or-cancel. People want delivery in seven days, of which three to four could be transport.”

Mort is one of four specialists overseeing the continuous re-engineering of Black & Decker’s manufacturing processes. Among other approaches including object-oriented programming, they are working on group technology, or the notion of creating mini-factories within a factory where a small group of people and machines work together. But instead of building these physically, he says, with Rhythm B&D should be able to create efficient virtual groups that can be reorganized in response to changing conditions.

Currently, Black & Decker’s systems (including Rhythm) classify labor by Job skill but without wage rates, and handle orders by date but without estimates of profitability or price differentials for timing.

Indeed, Intellection’s Sidhu says, most customers aren’t even close to knowing their costs, so the costs of most resources and trade-offs aren’t denominated in dollars, but on scales of preference generated by whoever sets up the system. The real goal, of course, is to keep everything close to capacity with enough flexibility to handle high-profit last-minute orders; calculating the expected value of those orders is the real way to determine the “cost” of any line. It’s an actuarial value which is itself determined by the intricacies and cross-dependencies of the schedule.

Likewise, many manufacturing operations pay little attention to the prices, let alone the profit margins, of the goods they produce. It’s all they can do simply to fulfill the delivery promises of their.sales operations.

Rhythm is written in C++, as one might expect, and represents all the elements as objects, with their constraints and other parameters defined as methods that interact with those of other objects. As shown above, users can interact with a wide set of customizable tools/views, including: 1. orders.editor; 2. problem window, showing capacity, materials and other problems/constraints; 3. resource load, showing projected load va. demand ‘which needs to be resolved when load exceeds capacity as in the example shown here; 4, order plan, showing the resources to fill a particular order; 5. constraint-anchored optimization (CAO) parameters, which let the user control the optimization process, resulting in … 6. post-CAO load, the plan for Assembly Line B; and 7. a Gantt chart.

Quote me a time

In the short term, Rhythm’s value comes when something unexpected happens: a new rush order, a machine breakdown, a quality or supply problem that requires a new schedule. It can also flag danger points, likely shortages and other items in response to continually updated information from the factory floor. Overall, it allows a manufacturer to increase responsiveness to customers while reducing costs. (It’s easy to be responsive if you don’t care how much you spend.)

Long run, it helps not just in allocating resources but in making investment decisions (the airline equivalent of selecting hubs and route-planning). “Rhythm has taken the guesswork out of our capital-equipment decisions,” says Caterpillar’s Jim Lock. “We can make simulations and figure out where we’ll get the most bang for the buck. Sometimes we found out we didn’t need a new machining center, for example; we needed additional tooling for an existing center.”

The long-run benefit will be the ability to reflect costs back to purchasers, whether in-house departments or ultimate customers. In turn, the prices customers are willing to pay will indicate the profitability of producing any particular line of goods. (Perhaps some items should be made only if a factory has excess capacity.)

Intellection over time

The company was founded in 1988 by Sanjiv Sidhu, now 35 years old, formerly with the Texas Instruments Artificial Intelligence Laboratory, where he worked on complex scheduling problems — not just in theory but for TI customers such as Ford and DuPont. The company’s staff of 30 is engineer-heavy and salesperson-light; most of the engineers have practical experience in scheduling problems. Sidhu grew up in Hyderabad, India, and came to the US in 1980. He earned a masters in chemical engineering at Oklahoma State and did PhD work at Case Western before being lured away by the opportunity to do work with real customers at TI.

So far, Intellection has won every account it has gone after — which indicates to Sidhu that the company must not be stretching itself enough. On the other hand, it probably couldn’t handle much more business than the fifteen clients it has now. Although the system itself is easy to use, it takes three to six months of set-up time with consulting from Intellection, since most companies don’t yet explicitly represent all the constraints, costs and other data that Rhythm requires to do its work. The new version should shorten the installation process by half, “We don’t want to be the bottleneck,” says Sidhu appropriately.

[1] Other customers include Ford, General Dynamics, Logan Aluminum, PPG, Eli Lilly, Heinz, Northrop and Solectron. Revenues last year were $2 million.

[2] How does it compare to She genetic-algorithm schedulers we described last year in Release 1.0, 5-92F You could replace some of Rhythm’s optimization algorithms with a genetic algorithm for scheduling, but a major parr of Rhythm’s contribution is the efficiency of its constraint-anchored search approach.

COPYRIGHT 1993 EDventure Holdings, Inc.

COPYRIGHT 2004 Gale Group