Predict Future Failures From Your Maintenance Records

Transcription

1 Predict Future Failures From Your Maintenance Records Presented by: Paul Barringer, P.E. Barringer & Associates, Inc. Reliability, Engineering, and Manufacturing Consultants Humble, Texas 77347, USA Phone: FAX: Barringer & Associates, Inc Crow/AMSAA Plots Most data fit a power law distribution which gives a straight line on log-log paper We had Duane plots, then Duane/AMSAA plots, and now Crow/AMSAA plots The methodology handles mixed failure modes and forecast future failures Results are simple and easy to use log-log plots of failures vs time easy to explain, easy to make, and easy to forecast failures Barringer & Associates, Inc

2 How Did This Get Started? T. P. Wright (1936) used the idea for learning curves in manufacturing The WWII War Production Board used Wright s techniques to forecast manpower James Duane of General Electric applied the concept to failure data for MTBF US Army Material Systems Analysis Activity wrote MIL-HDBK-189 and published new data in TR-652 (Growth Guide) download a copy from Barringer & Associates, Inc Figure 1 Note the time decrease of 20% for each unit Unit Cost (hrs) Unit Cost (hrs/unit) Aveg Cost (hrs) Y=X^( ) Units Produced Y=5.3268X^ Units Produced Y=5.3268X^( ) Units Produced T. P. Wright s Idea For Learning Curves Which Later Drives James Duane s Idea For Reliability Growth Plots Barringer & Associates, Inc

3 Table 1 Maintenance Interventions Month January February March April May June July August September October November 12 7 December 11 8 Total = TPM Began August 1995 Actual Failure Data From A Chemical Plant In Brazil Barringer & Associates, Inc Total ulative Maintenance Interventions Figure Reduction Of Pump Failures In A Brazilian Monovinyl-Chloride Plant By Use Of TPM Crow/rgr TPM Begins Aug 95 Before TPM Jan 95 Repair Reduction N=34.65t^0.947 with R^2=0.997 Big Improvement Barringer & Associates, Inc After TPM 6 Months (Total ulative) After TPM N=77.49t^0.529 with R^2=0.999 Engineering Starts TPM Jun 95 Lambda Beta r^ Slight Improvement Before TPM Updated Yr 2003 M01 D07 HPB

4 The Fearless Forecast Math N = λ*t β Before TPM N = 34.65*t After TPM N = 77.49*t N = cum number of failures λ = y-intercept at t=1 t = cum time β = slope of the line β < 1 improvement β = 1 no change β > 1 deterioration Forecast at t = 40 months N = 34.65* = 1140 Forecast at t = 40 months N = 77.49* = 545 = = 595 failures avoided Barringer & Associates, Inc The Fearless Monthly Forecast Maintenance Interventions Month '98 Fcst '99 Fcst January February March April May June July August September October November December Total = TPM Began August 1995 Barringer & Associates, Inc

5 Failures Represented By Monthly Maintenance Costs Table 3 Petroleum Refinery Department Maintenance Cost History For One Area $'s $'s $'s $'s Jan 31 $ 2, $ 4,146, $ 8,805, $ 13,627,145 Feb 59 $ 456, $ 4,450, $ 9,077, $ 14,076,446 Mar 90 $ 756, $ 4,846, $ 9,435, $ 14,275,526 Apr 120 $ 1,028, $ 5,129, $ 9,746, $ 14,537,284 May 151 $ 1,262, $ 5,673, $,135, $ 14,937,865 Jun 181 $ 1,540,1 547 $ 6,147, $,674, $ 14,732,077 Jul 212 $ 1,815, $ 6,896, $,957, $ 15,075,166 Aug 243 $ 2,121, $ 7,537, $ 11,420, $ 15,3,813 Sep 273 $ 2,769, $ 7,856,635 4 $ 11,932, $ 15,589,596 Oct 304 $ 3,047, $ 8,254, $ 12,857, $ 15,826,120 Nov 334 $ 3,360, $ 8,716, $ 13,402, $ 15,944,082 Dec 365 $ 3,748, $ 8,440, $ 13,214, $ 16,275,941 Circled Data Points Are Problems! The cumulative $ s decreases thus do not include this data point in the regression for β & λ. However note that the $ values continue into subsequent data points. Barringer & Associates, Inc Figure 3 Department Maintenance Data--1st Look Maintenance Costs (Total ulative US$) $,000,000 $1,000,000 Crow/rgr Zoom On This Region Trend Line Using All Data Lambda Beta r^ $,000 0 (Total ulative) Barringer & Associates, Inc. 2003

6 Figure 4 $ 20,000,000 Maintenance Cost (Total ulative US$) $ 18,000,000 $ 16,000,000 $ 14,000,000 $ 12,000,000 $,000, Department Maintenance Data-2 nd Look Crow/rgr Feb 2002 Cusp Month Savings $2,222,000 C$ = 3361t Announced TPM Program Started TPM Efforts Jan 2002 Dec 2002 C$ = 1839t Lambda Beta r^ (Total ulative) Barringer & Associates, Inc Table 4 Incidents Safety Record--Major Chemical Plant Incidents Incidents Incidents Incidents years Safety Incidents Are Failures! Long term average MTBF is 36/1 = 30.7 days/incident Barringer & Associates, Inc

7 Figure Crow/IEC1164 Safety Incidents Jan '94-Aug '02 Forecast Safety Incidents (Total ulative) 1 Unfavorable Trends 2-Aug Sep Nov Dec Feb Apr May Jul Sep Oct Dec Feb Mar May Lambda Beta Time (Total ulative ) Actual Forecast Barringer & Associates, Inc Table 5 Fearless Forecast Of Spills Spill Date Raw Data Crow/AMSAA Data Forecasts Failures Spill.. Predicted Between Events Spills By Old Spill Method 11/18/ /31/ /8/ /22/ /29/ /23/ /25/ New Method Savings 6/20/ /22/ // Failures In 899 This Is Good News! 8 Failures In 316 This Is The Datum Barringer & Associates, Inc

8 Figure 6 Crow/IEC1164 Spills Old Way Spills (Total ulative) Improvement Spill Reduction Lambda Beta Fit-P% <! >= Time (Total ulative ) Barringer & Associates, Inc Table 6 Spill Date Raw Data Crow/AMSAA Data Forecasted Failures Failures Missed New Spill.. Predicted Opportunities Between Method Events Spills By Old From Spill Savings Method Relapse 11/18/ /31/ /8/ /22/ /29/ /23/ /25/ /20/ /22/ // /16/ /7/ /12/ /29/ Bad News! Good News! 11/16/ /25/ Failures 5 3/25/ /1/ Coming 6 /28/ Faster! 7 7// /25/ /6/ /18/ Barringer & Associates, Inc

9 Figure 7 Crow/IEC1164 Spills Old Way Relapse Spills (Total ulative) Spill Reduction Improvement Retrench Lambda Beta Fit-P% <! >= >= Time (Total ulative ) We ve Lost The Way To Success! Barringer & Associates, Inc Table 7 Date Event Outage Event Description Between Event All Outages. Failures Forced Outages. Failures 2/1/1999 Planned Tie In /20/1999 Planned Tie In /24/1999 Forced Gas Line Outage /22/1999 Forced Animal Contact /9/1999 Planned Interconnect Energized /9/1999 Forced Switching Error /13/1999 Forced Tie Wrap Failure /13/1999 Forced Lightning Strike /3/1999 Forced Static Wire Short /6/1999 Forced Switch Failed //1999 Forced Not Logged /3/2000 Forced Cable Bond Fault /12/2000 Forced Underground Cable Fault /21/2000 Forced Bird Contact /11/2000 Forced Lightning Strike /7/2000 Forced Animal Contact /2/2000 Forced Animal Contact /12/2000 Forced High Winds /11/2001 Forced Not Logged /12/2001 Forced Not Logged /19/2001 Planned Tie In /7/2001 Forced Not Logged /22/2001 Forced Pole Damage /13/2001 Forced Interconnect Opened /16/2001 Forced Supplemental Power Out /6/2001 Forced Power Dip /12/2001 Forced Control Tripped /31/2001 Forced Power Dip /1/2001 Forced Power Dip /1/2002 Forced Steam Outage /15/2002 Forced Switching Error /18/2002 Forced Load Shedding Error /27/2002 Forced Water In Switch Gear /6/2002 Forced Generator Air Intake Frozen /3/2003 Forced UPS Failure Co-gen Failure Log: 31 Forced Outages In 1432 = 46.5 days per failure Barringer & Associates, Inc

10 Figure 8 Crow/IEC1164 Combined Cycle Co-Gen Plant Forced Outages (Total ulative) Forced Outages 1 0 Time (Total ulative ) Lambda Beta Fit-P% >= 5000 No Change Barringer & Associates, Inc Table 7 With Failure Forecast All Outages Forced Outages Date Event.. Event Description Between Outage Failures Failures Event 2/1/1999 Planned Tie In /20/1999 Planned Tie In /24/1999 Forced Gas Line Outage /22/1999 Forced Animal Contact /9/1999 Planned Interconnect Energized /9/1999 Forced Switching Error /13/1999 Forced Tie Wrap Failure /13/1999 Forced Lightning Strike /3/ /6/1999 Forced Forced Static Wire Short Switch Failed /27/2002 Forced Water In Switch Gear /6/2002 Forced Generator Air Intake Frozen /3/2003 Forced UPS Failure /23/2003 Forced Predicted /11/2003 Forced Predicted /27/2003 Forced Predicted More Failures /13/2003 Forced Predicted Predicted Into /28/2003 Forced Predicted The Future From /14/2003 Forced Predicted /29/2003 Forced Predicted Lack Of An /15/2004 Forced Predicted Improvement /2/2004 Forced Predicted Program (β 1)! 4/17/2004 Forced Predicted /3/2004 Forced Predicted Barringer & Associates, Inc

11 Summary Five actual datasets demonstrate straight lines of cum failures vs cum time Crow/AMSAA technology allows failure forecast of the mixed failure modes The forecast objective is to start decisive corrective action and prevent predicted failures Preventing failures requires proactive effort Preventing future failures saves $! Barringer & Associates, Inc