Modeling an SMT Line to Improve Throughput
One of the major challenges for an electronics assembly manufacturing engineer is determining how an SMT machine will impact throughput. Typically, an SMT equipment supplier will ask for a few (5-10) products to simulate the throughput capability of their machine. Unfortunately, if the engineer works in a high-mix, low-volume environment, he may need to know the impact of a new machine on 1,000 or more products. Currently, there are no simulation tools to effectively model this. This is confirmed in the 2015 IPC International Technology Roadmap for Electronics Interconnections, which states, "In order to better deal with the demands for increased interconnection density and respond to market demands for better return on capital investment in assembly equipment, there is a need within the manufacturing industry for continued improvement in tools and software for modeling and simulation. Needs in this area include better methods of load balancing and improved machine utilization. The tools for determining the balance on assembly lines will need to be flexible to handle the mix of assembly types that manufacturers now face."
Rockwell Automation partnered with Universal Instruments to develop a tool to model a large quantity of products and the impact of varying SMT line configurations. The information used for the modeling includes placements per panel and components placed per hour. With these tools, an electronics assembly plant can be analyzed to identify improvement opportunities and perform "what if" analysis to model impact of machine changes.
Goals for the SMT Line Model
1. Determine the right machine for the product mix.
2. Determine if products are running as fast as they should.
3. Determine if electronics assembly products are built on the optimal line configuration. This is crucial in plants with multiple line configurations.
Development of the SMT Machine Model
1. Discovery that machine cycle times were poor
After sample product simulations were run by Universal Instruments, it was discovered that observed cycle times were two to three times longer than simulated cycle times. This led to a focused effort to understand why. A kaizen event was held to map out the process and observe product builds. Several items that impacted the product cycle time were uncovered. These items were:
1. Component library placement speed slowed down.
2. Imbalance between placement beams/heads due to not having enough nozzles to pick and place the required component packages for the products.
3. Bypassed nozzles and spindles.
4. Large quantity of placements from a single component input.
5. Panel transfer rate into and out of the machine slowed down.
6. Poor optimization and component split between machines on an SMT line.
7. Operator variation in responding to the process.
The most significant item impacting cycle time was not having the necessary quantity of nozzles available for the mix of component packages for the products that the machine/line was building. To maximize flexibility to move products between lines, machines of the same type were equipped with a standard nozzle configuration. The nozzle configurations were changed only when a new component package was needed. To address this problem, a regular nozzle review was implemented to ensure the machines have sufficient nozzles available to optimize the machine programs.
Products were reviewed for the above issues. As items were addressed, the observed cycle times were reduced to align with the simulated cycle times.
2. Realization that cycle time does not represent SMT machine utilization
Cycle time represents how a product is running compared to a benchmark but does not reflect utilization of a machine based upon its throughput capability. For pick and place machines, throughput can be measured in components placed per hour (CPH).
Table 1. Sample of range of placements per panel to run IPC and manufacturer tests.
Manufacturers provide CPH specifications for SMT machines in two ways. The first method is what is often called "Maximum CPH", which represents the maximum speed the manufacturer was able to achieve and the second is based on "IPC 9850", which has CPH categorized by package type. The “placements per panel” required to run these tests are shown in Table 1.
The "IPC 9850" performance tests are useful to compare equipment models and manufacturers to each other, but they do not necessarily represent the products manufacturers are building. This complexity can be understood by comparing Table 1 to the sample product complexity of global product mix in Table 2.
Table 2. Sample range of placements per panel versus count of assemblies and forecasted panel volumes.
The idea of using simple regression to develop a model of “placements per panel” to CPH began to develop. This relationship was first studied using production history.
Machine Mathematical Model for CPH
A report was available that contained panels built and total time to build a work order. This report was used to calculate the average CPH per panel for an SMT machine model. A scatter plot with a smoother line was used to view the relationship between the variables for a machine model. The smoother line is a line fitted to the data to explore the potential relationships between two variables, without fitting a specific model, such as a regression line.
There is a relationship between “placements per panel” and CPH but there are points that do not follow the smoother curve. The other observation is that actual CPH values vary greatly compared to the specification value the manufacturer stated.
Since production data was used to model this relationship, all the problem areas outlined earlier represent part of the performance and added noise in the model. Another idea was to use generic product simulation data from the manufacturer. The product simulation information included:
1. Quantity of placements per panel
2. Simulated cycle time for a SMT machine
3. CPH (calculated)
This would filter out the noise from production and machine configuration issues and could then be used to establish a realistic CPH equation. With the simulated cycle time data, the relationship between “placements per panel” and CPH was then studied.
The scatter plot revealed a relationship between “placements per panel” and CPH. Using the Pearson Correlation Coefficient, the strength of the relationship is assessed. At 0.536 it is considered moderate and P-Value of 0.000 means the relationship is statistically significant. This indicates that “placements per panel” is a good predictor of CPH.
To read the full version of this article, which appeared in the May 2018 issue of SMT007 Magazine, click here.