Final Report. Volume 1: Final Design Report. Intelligent Machining & Process Control. April 22, 2009

Transcription

1 Final Report Volume 1: Final Design Report Intelligent Machining & Process Control April 22, 2009

2 MACHINELOGIC Intelligent Machining & Process Control Final Design Report William Dressel, Kevin Pham, Steven Stone, Sean Sullivan, & Phan Vu

3 Executive Summary Purpose To integrate the idea of Computer Aided Process Planning (CAPP) into the manufacturing process to help increase efficiency and reduce costs. To achieve this we will monitor tool wear, address machining stability issues, and record machining data to help determine optimal machining conditions. Key Requirements & Specifications o o Minimize work piece cost Determine tool cost Determine machining cost Balance process to minimize overall cost Provide feedback to digital manufacturing framework Develop data acquisition system Automate data and error logging Monitor machine stability: chatter detection Testing Results It is our conclusion that cost savings can be achieved through higher cutting speeds. Increases to cutting speed are limited to machining stability and stability can be monitored through a data acquisition system. Chatter is one main stability issue which can be detected with the use of a microphone to detect chatter frequencies. Business Case By increasing the spindle speed from 6338 in./min. to 6979 in./min. (110%) we can see the following cost reduction for the roughing process: Roughing Operation Optimized Spindle speed Feed rate Cycle time in Minutes Reduction in time in Minutes % reduction in cycle time Cost per part % Cost reduction 100% 100% % $ % 110% 100% % $ % Roughing Operation Optimized Initial Investment Investment per extra unit Units Ordered Total Investment Cost saved per part Cost saved per year Return on Investment Prototype investment $20,000 $8,893 1 $28,893 $0.58 $8, % Industry level investment(6 units) $106,000 $8,893 6 $159,358 $0.58 $49, % FINAL DESIGN REPORT i

4 Executive Summary Recommendations o o o o o o Test recommended settings, measure observed tool wear and confirm cost savings Find replacement for LabVIEW software as it is only intended for testing purposes Provide networked storage location for logged data Develop system for logging tool changes Relocate prototype to a more user friendly location Ability for DAQ board to interface with machine controller to control feed/speed rate FINAL DESIGN REPORT ii

5 CONTENTS 1. Introduction Introduction Purpose Scope of Report Tool Wear Analysis Background Information Tool Wear Analysis Tool Wear Analysis Results Prototype Design & Results Prototype Overview Prototype Architecture Prototype Testing and Results November Prototype Testing and Results February Prototype Testing and Results March Final Prototype Specifications and Recommendations Business Case Business Case Overview Selecting Optimum Setting Cost Reduction Return on Investment Conclusion In Summary Recommendations References...R-1 TABLES FIGURES 4.1 Optimal settings Cost reduction based on spindle speed Return on investment Results Comparison Overall system architecture Chatter detection using FFT Work piece cost diagram FINAL DESIGN REPORT ii

6 1. Introduction 1-1 Introduction Over the years, manufacturing companies have shown a great interest in automating the manufacturing process through the use of intelligent machines. An intelligent machine is a machine that is able to learn and adapt to changes within its system. Through the use of intelligent machines, manufacturing plants are able to increase efficiency, increase production, reduce costs, and reduce waste produced by human error. 1-2 Purpose The purpose of implementing intelligent machining in this project is multifaceted. The primary objective is to minimize the cost of the work piece, in this case a pipe coupling used in the oil industry, by optimizing the cost of machining and the cost of tooling through monitoring feed rate and its effect on tool wear. Additionally, we will address real time issues such as chatter and tool failure. Also, the intelligent machining system will integrate into a Computer Aided Process Planning (CAPP) system, providing useful feedback for the algorithms used to plan the machining process. 1-3 Scope of Report The report provides an overview of the final system design, it's performance in the field, a detailed description of the results obtained to date. The details and results from prototype testing are also reviewed. Finally, recommendations for future work are given. FINAL DESIGN REPORT 1-1

7 2. Tool Wear Analysis 2-1 Background Information From George Tlusty s Manufacturing Processes and Equipment book, the cost per part consists of three components; the fixed cost associated with the cost of the material, the machining cost, and the cost related to tool wear and tool change time. C part = C fixed + C mac ining + C tool _related Equation 1 The machining cost is $82.81 per hour. determined by the following equation: The cost related to tool wear and tool change time is C tool _related = t c r m +C te T t m Equation 2 where r m is the rate for machining a part ($82.81/hr), t m is the cycle time, C te is the cost per tool edge ($1 $2), t ch is the tool change time (1.5 minutes), and T is the tool life. The tool life equation is given as T = Cv p f r q Equation 3 where v is the cutting speed, fr is the feed rate, C is a constant, p is a constant, and q is a constant. To find the optimal feed and cutting speed, the cost per part equation first have to be rearranged into cutting speed, feed rates, and known parameters. Doing so will yield the equation shown below C part = c fixed + πd m L r m vf r + t c r m +C te Cv p +1 f r q+1 Equation 4 where D m is the average part diameter and L is the total linear length of cut. Taking the partial derivatives with respect to cutting speed and feed rate yield the following equations: v p fr q = Cr m p 1 (t c r m +C te ) Equation 5 FINAL DESIGN REPORT 2-1

8 v p fr q = Cr m q 1 (t c r m +C te ) Equation 6 Based on observation, equation 5 and equation 6 cannot be used to give unique finite values for the cutting speed and feed rate. However, since the feed rate and cutting speed settings are limited to stability, it can be seen that the minimum cost per part is attained when the feed rate is set to the highest value possible. From here, the optimal cutting speed can then be calculated as v opt = Cr m q p 1 t c r m +C te f max 1/p Equation Tool Wear Analysis As discussed in the previous section, the optimal cutting parameters is determined by setting the feed rate to be the at the boundary limit for stability, surface finish, and avoiding catastrophic tool failure. The optimal cutting speed is then determined from Equation 7. The process is run at these setting and in our case we monitor the process to ensure that it is stable: specifically we monitor the process of the onset of chatter based on a detailed system dynamic model, dynamic sensors, data acquisition system and signal processing. With the help of the staff at the manufacturing plant in Red Lion, Pennsylvania, a number of experiments were conducted to measure the tool wear on cutting inserts under a variety of feed and spindle speed settings. The tool life of the cutting inserts were then quantified by measuring the average flank wear width. The width was chosen because it is the most consistent form of wear according to Tlusty. From the experiment (see volume 3, p.7 for data), the following results were obtained in figure 2.2. Figure 2.1: Screenshots of the generated mesh from performing a linear regression on MATLAB. The red points on the graph are data points obtained from measuring the average flank wear width. FINAL DESIGN REPORT 2-2

9 The average flank wear width was measured using a computer interfaced optical microscope with image processing software. The optical microscope results were then compared with measurements obtained from a Wyko Profilometer (scanning white light interferometer) to verify the microscope's accuracy. The method for measuring average flank wear width was to determine the total area of the wear and divide the area by the flank wear width. Since the cutting inserts were used on a variety of parts, the flank wear had to be normalized with respect to the total amount of time the tool was used. To do that, the following equation was used: T = Flank_wear limit Flank_wear measured (cycle_time number_of_parts) After normalizing the flank wear with respect to minutes used, a linear regression was performed using the data obtained to get the best fit approximation for the tool life equation. Using MATLAB to perform a linear regression on the data obtained, the best fit equation for tool life is T = v q From the experimental data obtained, the highest setting for feed rate is 0.02 in/rev since no cutting inserts were shown to failed prior to reaching the targeted amount of parts. Substituting known values into the optimal cutting speed equation yields the following optimal cutting speed: v opt = Cr m p 1 t c r m + C te f max q 1/p = in min It is important to note that the optimal cutting speed value obtained above is not bounded by stability issue. From experience, occurrences of chatter are mostly likely to happen at this speed. However, the value obtained shows that the cost of machining outweighs the cost related to tool wear and tool change time. Using the data obtained, the optimal cutting speed is then recommended as 6970 in/min. 2-3 Tool Wear Analysis Results Based on calculations, that the cost of machining outweighs the cost related to tool wear and tool change time. The team goal was to find an optimal cutting speed and feed rate that is stable and will minimize cost per part. From the data collected for the NC50 Box Tx50Bo3-3, setting the cutting speed to 6970 in/min (330 rpm 110% override) and feed rates to 0.02 in/rev for 100 parts did not cause any FINAL DESIGN REPORT 2-3

10 inserts to fail prematurely. As a result, we are suggesting that the optimal cutting speed is 6970 in/min and the optimal feed rate is 0.02 in/rev. Under this setting the estimated tool life is 27.2 min. Under normal settings (300 rpm and 0.02 in/rev), the estimated tool life is min. FINAL DESIGN REPORT 2-4

11 3. Prototype Design & Results 3-1 Prototype Overview The main purpose of the prototype was to provide proof of concept for a data acquisition system that could record useful information about the machining process which could then be used to improve the process. The prototype included a computer, a data acquisition (DAQ) board, three sensors, and the Okuma LC40 Lathe. The sensors were mounted and connected to the LC40 lathe and the data acquisition board served as the interface between the sensors and the computer. The three sensors included a power sensor that recorded the power usage of the lathe, a microphone that recorded audio data from the lathe, and an accelerometer that recorded vibration data. The power data was recorded with the intention of using the data to determine cycle times for the machining processes, such as roughing and finishing, record average power usage for each process, and also to help predict tool failure. Both the audio and vibration data were recorded with the intention of using the data to detect chatter. The purpose of using both a microphone and an accelerometer to detect chatter was to make a comparison between the two types of sensors and determine which would provide a better implementation for detecting chatter. It was determined that the microphone provided a better implementation because it was cheaper, easier to mount, and its data had a significantly better signal to noise ratio. 3-2 Prototype Architecture Figure 1 shows an outline of our prototype architecture. The inputs included data from the sensors and the information input by the machine operator through the human machine interface. The data from the sensors is acquired through the DAQ board and then sent to the computer. The computer uses LabVIEW to analyze and record the data. The outputs included indicators for chatter on the human machine interface and the data that is logged. Fig. 3.1 Overall system architecture FINAL DESIGN REPORT 3-1

12 3-3 Prototype Testing and Results November 2008 A preliminary version of the prototype was tested on site in Red Lion, PA at the General Dynamics OTS factory located there in November The three sensors were installed on an Okuma LC40 lathe and some preliminary data was recorded. For this round of testing, the microphone was mounted outside of the lathe on the door, pointed in toward the work piece through a crack in the door. While this sufficed for preliminary data, it was determined that we needed to find a way to mount the microphone inside the lathe to prevent noise disturbances from outside the lathe from causing too much interference in the signal. There were also issues with mounting the accelerometer on the lathe. The issues involved finding a place to safely mount the accelerometer and keeping the accelerometer mounted permanently. In order for the data from the accelerometer to be useful, it needed to be mounted as closely to the work piece as possible. The closest location that was safe was on the turret housing. The ideal location would have been on the tip of the boring bar, as this is where the vibrations would be at their greatest. However, the accelerometer was too large to be mounted on the boring bar. Also, the boring bar and turret rotated, which would have caused issues with the wiring. Additionally, multiple accelerometers would have been required. On this trip, the accelerometer was tested on both the boring turret housing and the turning turret housing. The vibration data was much better on the boring turret housing. This is most likely because the vibrations from the boring bar would be greater than the vibrations from the turning mounts since the turning mounts were much stiffer. Mounting the accelerometer was also tested with both wax and a silicon epoxy. The wax would only last for about 3 parts, at which point the forces from the vibrations in the turret housing would knock the accelerometer loose. The epoxy was allowed to dry for approximately an hour, after which the accelerometer lasted for 17 parts. It was concluded from these tests that a more permanent method for mounting the accelerometer needed to be found. The power sensor was mounted inside the electrical cabinet of the lathe. The power data from these preliminary tests revealed that there was about a 7 volt spike when the spindle first accelerated to full speed at the beginning of the cycle, and a -5 volt spike when it was stopped at the end of the cycle. There was also a 4 volt spike when the turrets rotated between the roughing and finishing operations. These voltage spikes were used in later versions of the prototype to automatically begin and end data acquisition. 3-4 Prototype Testing and Results February 2009 Another trip to Red Lion was made in February 2009 to implement a newer version of the prototype and resolve some of the issues from the previous trip. This time the sensors were moved to an Okuma LU45 for testing the prototype. The microphone was again mounted outside the door while a better mounting location was researched. By the end of the trip it was decided that we would try to mount the microphone through a hole in the top of the lathe enclosure is in a water proof bag to prevent damage from the machining coolant. FINAL DESIGN REPORT 3-2

13 For this trip, a new method for mounting the accelerometer was tested. First an M6 stud was tapped so the accelerometer could screw into the stud. This stud was then screwed into the turning turret in one of the empty holes that is used to mount the tool insert holders. This method for mounting the accelerometer was a great improvement over wax or epoxy since it was far more permanent and did not fall off, but since it was mounted on a rotating part of the lathe, it was much riskier. Furthermore, there was not enough room to mount it on the boring turret, which would have been the preferred location. Ultimately, a couple of days after this trip, the accelerometer cable was ripped off the accelerometer from the rotating of the turret, which eliminated this as an option for mounting the accelerometer. The prototype employed in this trip used the voltage spikes to start and stop data acquisition automatically; however, sometimes the spikes were not detected by the software, so it was inconsistent. Eventually a solution was reached by making a digital connection to the limit switch on the lathe door. It was connected to our DAQ board and used programmatically to start data acquisition when the lathe doors shut, and to stop it when the lathe doors opened. This method was far more reliable and consistent, and is the final method used to automatically control data acquisition. On this trip, a computer was setup and connected to the sensors so that it could be left behind to acquire data automatically. There were a few problems with this setup. Primarily, the computer was setup at the end of the lathe at an inconvenient location for the operator to use. Also, it was difficult to train the operators how to use the GUI to record when tool changes were made and to indicate to the system when chatter occurred based on their personal observations. It was deemed critical to record these observations in order to correlate incidences of identified chatter by the algorithm with those that the machine operator identified. This correlation was used as the success metric. The system was able to collect data automatically for a few days after our trip, but the operators didn't always take the time to use the GUI to record useful information, so it was therefore mostly unsuccessful in its automatic data collection. 3-5 Prototype Testing and Results March 2009 Our final trip to Red Lion was made at the end of March On this trip the sensors were relocated back to the Okuma LC40 lathe based on the current production schedule. The microphone was covered in a water proof bag and successfully mounted in a hole drilled in the ceiling of the lathe. Having the microphone inside the lathe reduced the noise from outside disturbances. Also, a more suitable location for the accelerometer was found; screwed into a stud on the turret housing for the boring bars. Not only did this location not rotate, but it was also on the upper turret housing closer to the vibrations of interest from the boring bars. FINAL DESIGN REPORT 3-3

14 The prototype used on this trip was able to successfully detect chatter in real time using the microphone and accelerometer data. It did so using two methods, by monitoring the Fast-Fourier Transform of the data and by resampling the data at once per revolution of the work piece and monitoring the variance of the signal. After testing and comparing the chatter detection methods using both the accelerometer and microphone data, it was concluded that the most effective method was monitoring the Fast-Fourier Transform of the microphone data as this gave the best signal to noise ratio and was the easiest to establish the thresholds (Rivière et al, 2006). 3-6 Final Prototype Specifications and Recommendations Based on our tests, the final prototype suggested would not include an accelerometer, as it is expensive, hard to mount, and the microphone provides a better implementation to detect chatter. Figure 3.2 shows example data from the microphone and accelerometer that illustrates the better signal to noise ratio of the microphone. Note that the threshold magnitude for detecting chatter on the microphone is 20 times larger than the threshold for the accelerometer. This larger threshold is possible because signal of interest is significantly greater in magnitude than the noise, whereas the threshold on the accelerometer must be much smaller because the signal of interest doesn t have as great a magnitude. Furthermore, we were able to do all the necessary data analysis in LabVIEW and Matlab was not needed as originally expected. Fig. 3.2 Chatter detection using FFT FINAL DESIGN REPORT 3-4

15 Also, the location of the computer was still unsatisfactory. A suitable location next to the operator s workbench was found, but a table would have to be constructed for this purpose to fit the area available. The operators would need to be trained better in how to operate the GUI as this is crucial for gathering important data such as tool changes, chatter occurrences, and general information about the process. Also, if the data can be collected continuously with all tool changes properly logged, it might be possible to determine when a tool needs to be changed based on the power usage during a particular process (facing, roughing, or finishing). This work beyond our stated scope, but may be considered for future work. The basic concept behind predicting tool failure would be to look for a higher power usage than normal, which would be indicative of a dull tool insert. This would require for the average power usage for a good tool and a failing tool be known, and that the difference between the two would be enough to be easily detectable. Finally, it may be possible to automatically handle chatter if the DAQ board was wired into the override control for feed rate. One method of controlling chatter is suggested by operator experience at Red Lion. This method involves briefly increasing the feed rate whenever chatter was detected. This interruption is thought to interrupt the chatter. This would require that the output voltages necessary to change the feed rate to a desired setting be known, it would also require that the way the feed rate override works be known, and it might require a more expensive DAQ board capable of outputting specific analog signals. Alternatively, the method suggested by Tlusty of reducing the depth of cut should be tested as well. This method is described in Tlusty s Manufacturing Processes and Equipment in chapter 9 section 6: Chatter in Metal Cutting. FINAL DESIGN REPORT 3-5

16 4. Business Case 4-1 Business Case Overview The team s goal is to optimize the cost per part. The cost was calculated based on cycle time data, and the coupled tool wear data that was analyzed previously. The tool wear data was found by using the flank wear width measurements analyzed at different feed-rates and spindle-speeds. Then a best fit equation was calculated that quantifies how the changes in the feed-rate and spindle-speed will affect the flank wear width of the insert. Using this equation, it is possible to find the optimal feed-rate and spindle-speed needed in order to reduce the total cost of production per part. Fig. 4.1 Work piece cost diagram FINAL DESIGN REPORT 4-1

17 4-2 Selecting Optimal Setting Listed below are the nominal, optimal and suggested rates for both feed-rate and spindle-speed. The optimal values were derived through both the optimization formulas mentioned earlier and the tool wear analysis, but the values do not take into account stability issues, so it is suggested to use more reasonable values to test at first. The suggested spindle-speed is 110% the nominal value used at Red Lion for this work piece material and insert material. Nominal Optimal Suggested Nominal Cutting Speed 6338 in/min Optimal Cutting Speed 9066 in/min Suggested Cutting Speed 6972 in/min Nominal Feed Rate 0.02 in/rev Optimal Feed Rate 0.02 in/rev Suggested Feed Rate 0.02 in/rev Table 4.1 Optimal settings 4-3 Cost Reduction These suggested rates were used to calculate the new cycle times, and the cost per part. The two charts below have values based on both the optimization of the roughing operations only, and also the optimization of all operations. The experiment and tool wear analysis was done on the roughing operations only. The roughing operations were chosen because they are the longest of the operations. Using the values from the experiment and how the roughing cycle times were affected, we can extrapolate to the rest of the operations to find the values in the bottom of the two charts (All Operations Optimized). Roughing Operation Optimized Spindle speed Feed rate Cycle time in Minutes Reduction in time in Minutes % reduction in cycle time Cost per part % Cost reduction 100% 100% % $ % 110% 100% % $ % FINAL DESIGN REPORT 4-2

18 All Operations Optimized Spindle speed Feed rate Cycle time in Minutes Reduction in time in Minutes % reduction in cycle time Cost per part % Cost reduction 100% 100% % $ % 110% 100% % $ % Table 4.2 Cost reduction based on spindle speed 4-4 Return on Investment These next two charts use the cost reduction per part (found in the above charts), along with the amount invested in the system, in order to find a return on investment. These charts, like the two above are split in two based on which operations were optimized (using the suggested Spindle-speed and Feed-rate). Within each chart, there are two columns for investment options. The first is the prototype investment, which is what General Dynamics has already paid for the system (prototype and project costs). The other column is for Industry level investment, and includes running the system on six cells, rather than just one (includes an estimated, professional, product development cost). Based on the amount saved per year, and the return on investment, it is suggested to run the system on multiple cells to increase the savings: Roughing Operation Optimized Initial Investment Investment per extra unit Units Ordered Total Investment Cost saved per part Cost saved per year Return on Investment Prototype investment $20,000 $8,893 1 $28,893 $0.58 $8, % Industry level investment(6 units) $106,000 $8,893 6 $159,358 $0.58 $49, % All Operations Optimized Initial Investment Investment per extra Unit Units Ordered Total Investment Cost saved per part Cost saved per year Return on Investment Prototype investment $20,000 $8,893 1 $28,893 $0.81 $11, % Industry level investment(6 units) $106,000 $8,893 6 $159,358 $0.81 $69, % Table 4.3 Return on investment FINAL DESIGN REPORT 4-3

19 5. Conclusion 5-1 In Summary As a result of testing and experimentation, it has been discovered that cost savings can be achieved through higher cutting speeds. Increases to cutting speed are limited to machining stability and stability can be monitored through a data acquisition system. Chatter is one main stability issue which can be detected with the use of a microphone to detect chatter frequencies. Real-time monitoring of chatter along with data logging will provide the machinist and management with an archive of information to better determine machining parameters for determining future cost savings. 5-2 Recommendations Considerations for future development: 1. Find replacement for LabVIEW software as it is only intended for testing a. Possibly develop solution in house to minimize cost 2. Provide networked storage location for logged data 3. Develop system for logging tool changes 4. Relocate prototype to a more user friendly location 5. Ability for DAQ board to interface with machine controller to control feed/speed rate 6. Continue tool wear analysis a. Automate tool wear measuring process 7. Continue power data analysis 8. Automatically handle chatter through lathe control panel FINAL DESIGN REPORT 5-1

20 References E. Rivière, V. Stalon, O. Van den Abeele, E. Filippi, P. Dehombreux, "Chatter detection techniques using a microphone", Seventh national congress on theoretical and applied mechanics 2006, FPMs, Mons Delio T., Tlusty J., and Smith S., Use of audio signals for chatter detection and control, Journal of sound and vibration, vol. 114, pp , Manufacturing Processes and Equipment by J. (George) Tlusty, Prentice Hall, Upper River Saddle, NJ (2000). FINAL DESIGN REPORT R-1