Return to the Table of Contents.

7.5 Operations policies

Many more detailed questions still need to be answered:

• What plane arriving at an airport should be scheduled to land on which runway?
• In what order should the driver of a delivery truck make deliveries?
• Which patient waiting in an emergency room should be seen next by which medical professional?
• How should a log be cut into lumber at a lumber mill?
• In what order and using which hand should each operation be done in assembling a door handle and lock?

This section builds on two major concepts: the process matters and the IE sets policies.

First, the process by which work is done matters. The details of how workers do their jobs have small but cumulative effects that have big impact on efficiency, quality, and safety. An air traffic control system that consistently maintains required separations will be safer, even if each small lapse in maintaining required distances only slightly decreases safety. A delivery schedule that reduces travel time by just a few percent allows more deliveries in a day. A good method for sorting patients in an emergency room (such a method is called triage) can make sure that time is spent on the patients most in need of attention, thus improving quality of care. Cutting each log into a better combination of lumber of different dimensions can improve the company’s profit from each log. A well designed and tested process for assembling locks can reduce time, increase quality and reduce repetitive strain injuries. The process matters.

The second important concept is that the IE is involved in setting policies for how these decisions are made, not in actually making the decisions. Those policies are taught as Standard Operating Procedures and are incorporated into computer packages that support decision making. An IE can analyze different policies for choosing which plane should land next and determine which policies serve well in the long run. The policies can be encoded in a computer program that analyzes data on incoming planes, gate availability, and connecting flights and can display information visually and even recommend decisions to support decision making by the air traffic controller.

An IE can do the same for the delivery truck by creating a policy (perhaps a computer program) for laying out daily delivery routes. An IE can test the effectiveness of different rules for triage and implement the best rules as the ER policy. The IE can develop a combination of hardware and software to measure an incoming log and to compute the most profitable cutting pattern, given current market prices for different types of lumber. An IE can use various methods to devise the best assembly process, test the process in practice, and implement the process through training of workers. The IE devises the policies for how work is done.

IEs are in a very delicate position regarding workers. If you had 10 years of experience assembling locks, how would you feel if an IE just out of college told you how to do your job? IEs have to balance the fact that experienced workers often know how to do their jobs well with the fact that new ideas, new products, new equipment, or a bigger view of the system may mean that the IE just out of college does know a better way.

A new IE or an IE working with a new group moves slowly and builds credibility. Sometimes experienced workers are frustrated that their suggestions for improvement have been ignored. The experienced workers, if involved in a process improvement project from the start, may have ideas and interest in improving the process. Some companies have methods for involving workers in process improvement.

For example, quality circles, an idea borrowed from the Japanese, involve working groups of line workers who meet regularly to improve the production process in the area where they work. Sometimes such groups are called Kaizen teams, using the Japanese work for "change for improvement."

IEs can also reduce conflicts with workers by remembering that IEs don’t do the work. IEs set policies for the processes and then get out of the way. If the IE is engaged in doing the work, in repairing poorly done work, or in expediting work that has fallen behind, the IE should step back and figure out how to improve the process. If you ever find yourself reaching out to do some work that a worker should be doing, put your hands behind your back. Remember that the IE works on the system and the worker works in the system.

Constant improvement in the policies for operations can add up over time to considerable improvement in efficiency, quality, and safety. You may, as we have discussed already, be stuck with a poor layout, but you can still make a lot of improvements within that layout.

This focus on operations has risks. First, optimizing a small part of system may result in suboptimization for the larger system. Certain policies for landing planes may lead airline companies to change their schedules in ways that harm the overall functioning of the airport of or the air system as a whole.

Second, you can focus so closely on small issues that you miss the big issues; as the saying goes, you can’t see the forest for the trees. Perhaps the layout really is so inefficient, as compared to competitors, that the company will continue to lose money until the layout is redone. Optimizing small parts of a system while ignoring big problems that can destroy the system is sometimes described as rearranging the deck chairs on the Titanic.

The area of operations research (research on operations, called, more grammatically, operational research in Britain) has many fascinating ways to improve the efficiency, quality, and safety of operations, with primary emphasis on efficiency. Visual displays, analysis of data, and computer programs can help. Also, as we will see in Chapter 10), operations research can give us guidance about when the problem we need to solve would simply take too long and we should be satisfied with a good answer, even if it may not be the best answer. The field of operations research solves many interesting operations problems. I will now describe some of the problems and approaches used in scheduling and in routing.

Scheduling

Even a simple scheduling problem can be difficult, but a Gantt chart can help you visually evaluate schedules.

Example. Two jobs are waiting to be processed at two machines, A and B. Each job must be processed first on A and then on B. We want to minimize the total time to complete both jobs. This table show the processing time of each job on each machine in minutes.

The Gantt chart below shows that the total completion time is 110 minutes if we start with job 1. [do Gantt chart]

This second Gantt chart below shows that the total completion time is 130 minutes if we start with job 2. Since we wanted to complete both jobs as quickly as possible, we should start with job 1. The total time to complete both jobs (called the makespan) will be 110 minutes. [do Gantt chart]

In this case, the best decision wasn’t hard to find and perhaps you can see that we did job 1 first because it required less time on machine A than did Job 2. That decision policy is called Shortest Processing Time or SPT. Is it always a good policy? It works well for 2 jobs on 2 machines, but not when we have more jobs, as this example shows.

If we use SPT to select which job to run on machine A whenever we have a choice, we would do the jobs in the sequence 1, 2, 3, and this Gantt chart shows that this sequence gives a makespan of 200 minutes. [Do Ganntt chart]

However, this Gantt chart shows that the sequence 3, 2, 1 gives a makespan of 180. The sequence 3,2,1 is, in fact, the best, as you could confirm by checking all 6 possible sequences. [Do Ganntt chart]

If you are scheduling n jobs on 2 machines, there are n! possible sequences (you have n choices for the first job, n-1 for the 2nd job, n-2 for the third job, and so forth). Checking all those sequences would take quite a while if n is large. Johnson’s algorithm is guaranteed to give us the best sequence of n jobs on 2 machines:

1. Select the smallest processing time for all jobs on both machines.
2. a. If the time selected is for Job j on machine A, schedule job j next on machine A. Cross out Job j.
   b. If the time selected is for Job j on machine b, schedule job j next on machine b, working backward in time. Cross out Job j.
3. Repeat 1 and 2 until all jobs have been crossed out.

Try this algorithm on the three-job data above and confirm that you get the sequence 3, 2, 1.

This discussion leads again to our important results: local optimization (for example, schedule next on this machine the job with the smallest processing time) may not lead to the best overall system performance. Therefore, someone has to set policies for scheduling across the whole system.

If scheduling three jobs on two machines can be tricky, imagine how complicated bigger scheduling problems can be? Johnson’s algorithm gives us the best sequence for n jobs on two machines and a similar algorithm works in some circumstances for sequencing n jobs on three machines, but then we don’t have algorithms that work in general. You also can’t examine all possible sequences to find the best one, so you may need to be satisfied with a good sequence, even if it is not the best. Operations research analysts have found some generally good approaches, depending on your goal:

• SPT, or Shortest Processing Time. At each machine, select from all the jobs waiting for processing the job with the Shortest Processing Time on that machine. SPT scheduling on a single machine minimizes the average time all jobs spend in the system.
• FCFS, or first come first served. At each machine, select from all the jobs waiting for processing the job that arrived first.
• EDD, or earliest due date. At each machine, select from all the jobs waiting for processing the job that has the earlier due date, that is, the date that it is due with the customer.
• Critical ratio. For each job waiting for processing at a machine, compute the ratio of (total processing time left on that job)/(time left until the job is due) and select the job with the largest critical ratio.
• Lowest possible slack. For each job waiting for processing at a machine, compute the slack, that is, the time until the job is due minus the total processing time left on the job, and select the job with the smallest slack.

Each of these approaches has strengths and weaknesses and much research has been done on which approach to use in which circumstances.

Another possible approach to scheduling focuses on bottlenecks. Recall that a bottleneck is the resource with the smallest capacity. The flow rate through the bottleneck limits the flow rate through the entire system, so the goal of scheduling should be to keep the bottleneck busy all the time. Time wasted at the bottleneck is time wasted for the entire system. Staffing should be arranged so the bottleneck resource is running during breaks and lunches. A new job is scheduled into an available time slot on the bottleneck machine and then is scheduled backward and forward in time to the other machines.

Eliyahu Goldratt has generalized this focus on bottlenecks into the Theory of Constraints (TOC). An organization's performance is limited by its weakest link, so identify that link and increase its performance; the result will be an increase in performance for the entire organization. Repeat. The book The Goal by Eliyahu M. Goldratt and Jeff Cox, is a novel that introduces Goldratt's ideas.

Actual scheduling problems can be very complicated. Hsu et al. describe the requirements for scheduling road test on vehicles each day at the General Motors Development Center in Kapuskasing, in northern Ontario. Vehicles must cool off between tests, must be road tested in groups, and must have a minimum number of escort vehicles since the tests are done on public roads. Also, drivers have a maximum number of hours they can work and have required breaks.

Since such situations have unique requirements which may change over time, an IE usually tries to devise clever heuristics; a heuristic for solving a problem is a solution method that gives a good, if not optimal, solution. Clever heuristics can be developed that take advantage of the structure of the problem in that situation. A heuristic can be implemented in a computer program that asks the human user to input data, constraints, and weights on different objectives, and that then suggests a solution for the human to adjust. Often a “what if” capability lets the human try out and compare different solutions.

The General Motors authors tried using a standard optimization package (CPLEX) but the resulting problem was too big to be solved in a reasonable amount of time. They devised a heuristic to develop a number of feasible schedules and pick the best among them. The results were a reduction in time to make a schedule from nearly four hours to a few minutes, an increase of more than double in the average daily throughput, a reduction in corporate warranty costs in the millions of dollars due to the ability to perform more tests, and an improvement in employee satisfaction because of the improved quality of schedules.

Scheduling problems also arise with scheduling workers. In the airline industry, among others, workers submit requests for schedules and the company would like to satisfy as many requests as possible. In addition, safety rules may place constraints on lengths of shifts and time between shifts.

Gordon and Erkut describe a spreadsheet they designed to devise a schedule for volunteers at the Edmonton Folk Festival in 2003. Even scheduling a team with only 32 members became very complicated in an attempt to satisfy as many preferences by volunteers as possible. Edmonton hosts many festivals each summer and the competition for volunteers is fierce. The Edmonton project was on a small scale and took two months to implement.

Continental Airlines uses an optimization program to construct schedules for pilot training.

“A dedicated team of operations research and software professionals, including specialists in large-scale optimization, databases, C++, Visual Basic, and quality assurance from Navitaire as well as manpower planners, training schedulers, managers, and information technology professionals from Continental Airlines, worked together on this project for over two years.”

The project has paid off

“Overall, Continental has estimated savings in excess of $10 million annually from using the system to create training plans. “ (Yu et al., page 260)

Routing

The driver who has to deliver packages wants to pick the best route, that is, the sequence in which to deliver packages. Consider this problem, an example of what is called the Traveling Salesman Problem, where the driver has to make loop starting at A and visiting B, C, and D before returning to A. The table gives the distances between each site that must be visited.

In this case, the distances and the matrix are symmetric. In this small problem, you can try all the different orders and find the smallest (it has length 18), but a large problem can be very difficult and we may need to settle for a sequence that is good even if we can’t show that it is the best sequence.

A very clever approach to the Traveling Salesman Problem that can be easily applied is the Meals On Wheels algorithm, which uses space filling curves, as explained here.

Blakeley et al. created a decision support system to help supervisors in over 250 offices of the Schindler Elevator Corporation, Inc., schedule and create routes for maintenance visits. The project took approximately $1 million to implement, but reduced the route building process from several weeks in each office to a few hours. The total savings are estimated to be over $1 million annually.

Warehouses and distribution centers use workers to retrieve items based on customer orders. While much research has been done on how to create optimal routes for order picking, Dekker et al. describe a warehouse in the Netherlands that had unique features (for example, two floors) and unique requirements (for example, breakable items must be picked last). By analyzing various heuristics for generating routes and by analyzing data on the popularity of products (10% of products account for 70% of the total picks), the authors reallocated storage locations to improve the routes selected by the heuristics. However, since travel time accounted for only about 13% of the time to pick an order, the total time was reduced by only 4%. Other changes were used to reduce the picking time (for example, using scanners with a quicker response time). All the changes enabled a reduction in the number of pickers from 20 people to 12 to 15 people, depending on the level of demand.

These examples show just a few of the ways operations research can be applied to improve operations. In all of these examples, I have assumed that all the data for the problem are known in advance with certainty, but the real world involves uncertainty. We use probability as the mathematical language to describe uncertainty. Operations research is discussed more in Chapter 10.