Availability Modeling for the Application of Manufacturing Equipment Aron Brall, Landis Gardner, a UNOVA Company, Waynesboro Key Words: Reliability Model, Availability Model, Manufacturing Reliability, Industrial Reliability, Process Reliability SUMMARY AND CONCLUSIONS In this paper, I have shown that two significant areas of manufacturing availability are not addressed with traditional reliability modeling. First, manufacturing goals quality parts at the required production rate - can be met despite the fact that the system did not meet its uptime requirements. Second, process failures can contribute more to manufacturing downtime than actual hardware failures. Proper modeling of manufacturing systems to include all sources of downtime, whether hardware, software, human, or process as well as the production rates, will provide a more accurate depiction of operation. Methods have been developed by others to model software and human reliability. The approach paper details a model for including process reliability and production capacity. The resultant analysis provides a method to help assure that reliability improvement efforts are focused on the primary root causes of downtime in many cases process problems and not more costly and sometimes technologically infeasible hardware improvement. The analysis also points to addressing the production capabilities so that the operational goal for the manufacturing equipment is not missed. 1. INTRODUCTION Before discussing manufacturing reliability, it is necessary to establish the terminology of manufacturing. The following list defines the special terms used in this paper and some specific definitions of other common terms. All terms not specifically defined are in accordance with references 1, 2, and 3. For a more complete list, see reference 1 and the Bibliography contained therein. Nomenclature List Blocked an event when manufacturing equipment cannot produce items due to the output conveyor being full. Starved an event when manufacturing equipment cannot produce items due to the output conveyor being empty. Downtime the elapsed time when a machine is not capable of operating to specifications. Total Downtime includes Scheduled Downtime and Unscheduled Downtime (See figure 1.1) Uptime total time that a machine is on-line (powered up) and capable of meeting functional requirements.. Uptime includes Gross Operating Time and Non-operating Time. (See figure 1.1) Yield the percent of products meeting quality standards produced by the machinery or equipment. Overall Equipment Effectiveness (OEE) the product of availability, performance efficiency and yield. Performance Efficiency the ratio of the actual production rate to the specified production rate. Six Sigma A statistical methodology that tries to control process variation that is s-normally distributed such that the process tolerances are set at the six sigma points on the Gaussian distribution. Coolant a fluid used in manufacturing operations to lubricate machining processes, may be water based or oil based. In traditional Availability Modeling, a system is evaluated for the occurrence of hardware and software failures and the maintenance time required to restore the system to proper operation. The maintenance modeling can include logistics times to obtain men and materials and the actual repair times. Sometimes human failures are included in the analysis, and sometimes resource limitations are used in the modeling. When manufacturing equipment is applied to a production line, there are additional sources of unavailability than hardware or software failures. Some of these will be addressed in the process model presented here. When parallel machines are required to produce at a specified production rate, there is production of parts even when the system is technically in a failed condition because one or more machines are down. This effect will also be addressed in the process model that is presented. This paper will discuss the modeling of availability including the effects of process failures and system capacity. Process failures include the inherent variability within specification of manufactured items; the variation, within specification, of the various manufacturing operations, the effects of repair within specification, maintenance induced variation, and failure to perform or ineffectiveness of Preventive and/or Predictive Maintenance. issues are related to the maximum and nominal processing rates for equipment. While the basic availability requirement for manufacturing equipment is usually specified as percent uptime or OEE for individual machines, the quantity of finished product produced as a percentage of product required is the desired outcome that plant management usually measures (i.e., there is Copyright IEEE 2002 RAMS Conference Page 1
a requirement for 400,000 engines and only 390,000 engines were produced yields 97.5%). Monte Carlo simulation with capacity analysis provides a tool for predicting the ability of a machine or group of machines to produce the required quantity of parts. Manufacturing equipment has been traditionally sized based on existing production lines and simulation studies. The R&M characteristics used in these studies have been simplistic and opportunities to assure throughput for minimal, or sometimes no additional expense have been missed. The simulation studies presented in this paper show the impact on a particular operation or series of operations due to failures of the machinery (hardware and software) as well as failures of the process. The models used are empirical and based on actual field data collected over a number of studies and represent thousands of machine operating hours. The impact of modifications to the maintenance methodology, process robustness, and preventive maintenance schemes is evaluated and compared to the baseline. The capacity values are from actual requirements in an automotive plant. Simultaneous Engineering, a process for developing the manufacturing process during the product design phase, was implemented to assure a robust manufacturing process. However, by not utilizing R&M models that address the impact of various process choices, substantial process improvement opportunities have been left on the table. The use of an R&M model that addresses process as well as hardware failures and capacity analysis, can provide greater insight into causes of downtime and provide a production line with sufficient capacity to meet required production goals 2. MANUFACTURING APPLICATIONS Manufacturing reliability has significant differences from product reliability. In most manufacturing applications, machinery is used at flow rates that are less than the maximum design throughput. This allows for some downtime (scheduled and unscheduled) without the loss of total required volume. In many situations modeling reliability without addressing capacity, results in the system being overdesigned, or the rejection of solutions that can address the capacity requirement. This is particularly true when multiple machines are performing the same operation in parallel. 3. TRADITIONAL MANUFACTURING MODELS Traditional manufacturing models assign failures and associated repair times to the machine or section of a machine that reported the problem. These models disregard process failures as independent of hardware and software failures. These models also don t consider capacity. If any machine is down, the model treats the operation as being down and not meeting the specified requirements. If an electrical component is replaced, the electrical system is charged with the failure; if a build-up of metal cuttings causes the problem, the machine tooling may be charged. The problem is that when the R&M are modeled, the failures are always considered machine failures (including software), not process failures. Problem solving to improve up-time performance rarely or never addresses the process. The process gets addressed during development of the manufacturing line, then rarely once the machines are designed and delivered. Failures (downtime occurrences) of the process are not addressed at all. Failures for out of tolerance parts are addressed, but usually as machinery failures. It should be noted that in Japan, process failures (short stoppages of 5 to 10 minutes) are not included in R&M calculations at all. In many American plants, these short stoppages are used, particularly by the machinery manufacturers, to push the value of mean time to restore calculations down, without addressing the actual time to repair a hardware failure. Figure 3.1 is an R&M model for a traditional manufacturing application. Table 3.1 shows the input values used to run the model these values ignore all process failures. Tables 3.2 and 3.3 show the results for R&M after 100 simulations of 96 hours run time (a typical 6-day, two shift week). The system availability is 45%, and the production is only 8,614 camshafts. This occurs, even though each machine has an availability of >91%. The production efficiency of 8,614 units out of a planned 19,200 camshafts (200 units per hour X 96 hours) is 45%, the same as the system availability. 4. PROCESS FAILURES Process failures occur when the production line stops without any machinery failure or when a defective item is produced. I became aware of these failures while evaluating machine performance studies performed in different manufacturing plants manufacturing similar items on nearly identical machines. Numerous stoppages occurred that did not require the replacement or repair of a machinery component. Some occurred due to the buildup of metal shavings, swarf, and turnings from the manufacturing operations, some from the build-up of coolant deposits Production lines stop, without failure of the machinery due to variation in all the elements of the process including part touching elements of the machinery, a sort of dynamic tolerance stack-up. While each individual machine is designed and qualified to the specified process tolerances and Process FMEA is performed to address the interactions of operations, the processes in many instances lack robustness at the tolerance extremes. Without a tool to quantify the impact of these failures, resources are not applied to reducing their occurrence. Defective items can be produced when the process is performing as designed due to the statistical variation inherent in the process. For example, due to the inherent variation in the process, there is a finite probability that a properly working machine can produce a defective item. When the process tolerances are established as the three sigma points 3 parts in 10,000 can be out of specification for each characteristic. With large numbers of characteristics and large volumes of parts, the probability of a defective part due to process variation is very high. Even when the process is functioning at six-sigma process control levels, out of specification parts can occur.. As an example of this, consider a V-6 engine crankshaft. There are as many as 185 Copyright IEEE 2002 RAMS Conference Page 2
CL M M Mains Grinder CL Cam Lobe Grinder Traditional Manufacturing Block Diagram Figure 3.1 CL CLP M MP M Mains Grinder MP Mains Process CL Cam Lobe Grinder CLP Cam Lobe Process Process Manufacturing Block Diagram Figure 5.1 Copyright IEEE 2002 RAMS Conference Page 3
characteristics measured after processing a crankshaft. All of these characteristics are statistically independent of each other within the specified tolerances that is the changes in each of the characteristics do not correlate with one another. Even if each characteristic is controlled to six sigma around the nominal value (3 parts per billion), the result for all characteristics is 555 parts per billion. In a plant producing 1,000,000 engines a year that yields a probability of 55.5% that a defective part would be produced. In reality, the processes in the best plants are controlled to six sigma about the process mean and in a typical plant to four sigma about the nominal mean. This results in the probability of defective parts being nearly 100% - even though nothing has failed in the manufacturing machinery. 5. MODELING WITH PROCESS FAILURES Process failures are modeled as separate blocks in the model. Each Process Failure block is then associated with a machine (see figure 5.1). The repair times were divided to associate the typically shorter times (about 6 minutes mean) with the process failures, and the longer repair times (about 2 hours mean) with the machinery (see table 5.1 for input parameters). The failure rates for the process failures were developed from actual factory data and are given in Table 5.1. These are typical of automotive plants in operation today. The hardware failure rates were developed from the same set of data. In addition, machines were allowed to stay in production when other machines were down as long as the theoretical production rates were met. The results for 100 simulation runs are shown in Tables 5.2 and 5.3. They show that the system is capable of producing 19,157 of the planned 19,200 camshafts, for a production efficiency of 99.8%. Further analysis showed that allowing the model to run below the theoretical production rates produced 19,172 cam shafts, and adding 1 extra camshaft per hour at the input to the system (201/hour) would yield 19,274 camshafts or greater than the desired production. In addition, the individual machine availabilities were actually greater despite the overall failure rate of the machines being identical (see Table 5.3 for the simulation results). This is traceable to the short repair times being properly assigned to the high failure rate process failures and the long repair times being assigned to the lower failure rate hardware failures. 6. IMPROVING MANUFACTURING THROUGHPUT The application of process failure modeling with capacity analysis can provide dramatic insight into the manufacturing capabilities of machinery and equipment. The use of very conservative failure rate and repair data, that is, unedited plant data can be applied to identify material handling schemes to take advantage of parallel manufacturing operations. The process models used in this paper made one assumption about material handling that one machine in a down state did not impact the ability of another machine to receive or deliver parts. Using this and similar tools can allow a manufacturing engineer to perform trade off analysis during simultaneous engineering to determine combinations of parallel and series manufacturing operations that assure throughput that meets the required production goals of the plant. Additionally, manufacturing plants that are not meeting required production goals, can be analyzed to determine if operational changes, material handling changes, or other modifications will improve throughput. ACKNOWLEDGMENT I wish to thank Larry Wolfe formerly of ARINC for providing a beta copy of RAPTOR 6.0 which proved invaluable in developing the capacity estimates for complex manufacturing systems used in this paper. REFERENCES 1. Society of Automotive Engineers (SAE), Reliability & Maintainability Guideline for Manufacturing Machinery and Equipment, Second Edition, Warrendale, PA, August 1999. BIOGRAPHY Aron Brall Landis Gardner, a Division of UNOVA Industrial Automation Systems, Inc. 20 East Sixth Street Waynesboro, Pennsylvania 17268 USA E-mail: abrall@landisgardner.com Aron Brall is Vice President of Quality at Landis Gardner, a Division of UNOVA Industrial Automation Systems, Inc. He has had several positions there in Product Assurance since 1991. Prior to that he worked 12 years for the Amecom Division of Litton Systems as a Systems Effectiveness Project Engineer. Out of thirty-four years professional experience, twenty-seven have been in Reliability and Product Assurance. He received a BS in Electrical Engineering in 1967 from Columbia University, NY, NY, and an MBA in 1987 from Loyola College, Baltimore, MD. He is a senior member of ASQ and SME and a member of the IEEE, SRE, and SAE. He is an ASQ Certified Reliability Engineer. He is a contributing member of the committees that prepared the initial and revised editions of SAE M-110, Reliability and Maintainability Guideline for Manufacturing Machinery and Equipment. He is also a member of the RAMS 2002 Management Committee Copyright IEEE 2002 RAMS Conference Page 4
Block Name Failure θ (hours) Repair µ σ Maximum Nominal Input to System N/A N/A N/A N/A N/A 200 200 Mains Grinder Exponential 13.370 Lognormal -1.5034 1.8226 75 50 Cam Lobe Grinder Exponential 19.6106 Lognormal -1.5034 1.8226 25 20 Input Data for Traditional Manufacturing Model Table 3-1 Parameter Minimum Mean Maximum Std Dev Units Produced 816 8614 14667 2791 System Availability 4.25% 44.87% 76.39% 14.54% MTBMA (hours) 0.064 0.604 1.150 0.225 MRT (hours) 0.365 0.940 1.911 0.318 Output Summary Table for Traditional Manufacturing Model Table 3-2 Machine Availability Main Grinder 1 91.24% Main Grinder 2 94.03% Main Grinder 3 93.78% Main Grinder 4 92.17% Cam Lobe Grinder 1 96.65% Cam Lobe Grinder 2 95.07% Cam Lobe Grinder 3 95.53% Cam Lobe Grinder 4 96.62% Cam Lobe Grinder 5 93.94% Cam Lobe Grinder 6 95.31% Cam Lobe Grinder 7 97.45% Cam Lobe Grinder 8 92.42% Cam Lobe Grinder 9 92.38% Cam Lobe Grinder 10 94.97 Output Table for Grinder Availability Using Traditional Model Table 3-3 Block Name Failure θ (hours) Repair µ σ Maximum Nominal Input to System N/A N/A N/A N/A N/A 200 200 Mains Grinder Exponential 122.000 Lognormal -0.206812.832555 75 50 Mains Process Exponential 15.015 Lognormal -2.204301 1.227501 N/A N/A Cam Lobe Grinder Exponential 213.000 Lognormal -0.084209.832555 25 20 Cam Lobe Process Exponential 21.599 Lognormal -2.204301 1.227501 N/A N/A Input Data for Process Manufacturing Model Table 5-1 Parameter Minimum Mean Maximum Std Dev Units Produced 18855 19157 19200 53 System Availability 98.20% 99.77% 100% 0.28% MTBMA (hours) 0.975 1.301 1.920 0.164 MRT (hours) 0.201 0.305 0.460 0.060 Output Summary Table for Process Manufacturing Model Table 5-2 Copyright IEEE 2002 RAMS Conference Page 5
Machine Machine + Process Availability Main Grinder 1 97.80% Main Grinder 2 97.65% Main Grinder 3 97.60% Main Grinder 4 97.80% Cam Lobe Grinder 1 98.30% Cam Lobe Grinder 2 98.41% Cam Lobe Grinder 3 98.04% Cam Lobe Grinder 4 98.21% Cam Lobe Grinder 5 98.32% Cam Lobe Grinder 6 98.07% Cam Lobe Grinder 7 98.30% Cam Lobe Grinder 8 98.12% Cam Lobe Grinder 9 98.37% Cam Lobe Grinder 10 98.52% Output Table for Grinder Availability Using Process Model Table 5-3 Copyright IEEE 2002 RAMS Conference Page 6