|
Process Reliability
Punch List |
What is the issue for process reliability?-
- Most businesses have too much variability in their output.
- Variable output fails to provide a steady stream of income for the business.
- Quantity of product produced is a precursor for money.
- Reasons for variable output are many but lack categorization for corrective action—this is where process reliability tools help.
- Production personnel expect make-up of reduced output will occur the next day. Tomorrow never comes for make up and the business suffers from lack of revenue.
- The issue for process reliability is to quantify how the plant is performing and identify losses for corrective action.
- What’s the
reliability of your process and how can you show it?
How to resolve
the issue of process output variability?
- Use the daily production output to demonstrate, on one page, widely varying production output; and categorize the losses.
- Quantify the hidden factory losses on one side of one sheet of paper along with nameplate (entitlement) capability.
- Build and implement a Pareto corrective action plan for reducing output variability, and eliminate the hidden factory causing the variability.
- Identify categories of problems needing corrective action: reliability issues and efficiency/utilization issues.
- Demonstrate consistently high production output with small variability for stable cash flows with increased profits by eliminating waste and inefficiencies.
Typical findings
from process reliability analysis:
- Most operations have few clues about their large variability in output and consider themselves “best in class” and unbeatable.
- Most operations think their real problem is due to maintenance problems (an issue with “things”) when in fact more operations suffer from management driven issues influencing efficiency and utilization of the production process.
- Most processes lack consistently high output of product to yield a constant profit stream to the company.
A) Salaried employees want a constant stream of income from a monthly paycheck—failure to achieve a steady monthly paycheck is untenable.
B) Salaried employees strongly favoring a constant income allow the money machine (the process) to have enormous output swings, which fail to supply the business with a constantly high income—failure to achieve a steady business result is undesirable and upsetting to business mangers. - Most businesses can improve their output but it requires making a change to get a change by altering the status quo of people who design and control the process.
- About 65% of all processes have major management driven issues of efficiency and utilization—this category usually exceeds equipment issues.
- Improving the process for reducing output variability
from the plant is mandatory for achieving an excellent “world-class”
production plant.
The
punch list for making a process reliability plot for analysis:
- Gather 365 consecutive days of output from the process (see the attached spreadsheet for a working example).
- In column A of an Excel spreadsheet input the calendar date.
- In column B of the spreadsheet, input the daily output
- In column C of the spreadsheet, input the product grade—if mixed grades, input the preponderant product grade.
- In column D of the spreadsheet, write and “IF” statement “=if(b3=0,0.1,b3)” which will convert the zero values into some small value, say 0.1, so the zero value can be observed on a logarithmic scale.
- Copy the 365 days of production from column D and paste the data into WinSMITH Weibull. The resulting plot is shown in Figure 1
which requires the use of good engineering judgment to obtain a valid analysis—don’t blindly follow the punch list for every case!
|
Figure 1: 366 Days Of Production In A Weibull Plot |
|
|
- Zoom in on the steep portion of the curve by use of the
magnifying glass icon and select on the plot by drawing a box around the
section of interest. Curve steepness hides many details on the logarithmic scale to produce Figure 2.
|
Figure 2: Zoom On Step Section
Of The Curve |
|
|
- The steep portion of the curve shows three distinct zones of roughly parallel features with a few data points as transition between the zones—what causes the “stair steps”?
- Under the mixture/process reliability icon, fit the demonstrated production line to the upper right hand reaches of the plot in Figure 2. Use the production line/plot point fit menu. Draw a box around the upper data from 8.5% toward the right hand corner. Figure 3 shows the results where eta value is a key performance indicator for process capability. Eta provides a single point estimate for demonstrated output.
|
Figure 3: Fit The Demonstrated Production Line |
|
|
- Under the mixture/process reliability icon, set the reliability point at 8.5%. Also choose change points at 44.8%, 61%, 64%, 91.8%, and 97% using the menu item for setting at a value.
- Under the mixture/process reliability icon, click on
the green check mark or use the menu item to activate the analysis to find
the loss gaps shown in
Figure 4 showing why cusps for change points are important.
|
Figure 4: Sum Gaps To The Left Of The Demonstrated Production Line |
|
Total Production = 868,199.1 (Metric Tons) Production Line = Eta 2,775.538, Beta 16.76324 Nameplate Line = None Total Reliability (8.5-100%): Loss = 116,126 (Metric Tons) ßEqual
to 116,126/2,775.538 = 42 days Process Reliability (%) = 8.5 8.5-11.8%: Loss = 1,187 (Metric Tons) ßTransition zone 11.8-61%: Loss = 25,897 (Metric Tons) 61-64%: Loss = 2,981 (Metric Tons) ßTransition zone 64-91.8%: Loss = 35,759 (Metric Tons) 91.8-97%: Loss = 26,787 (Metric Tons) ßTransition zone 97-100%: Loss = 23,515 (Metric Tons) Without a nameplate line all losses look
like reliability problems! |
- The three transition zones represent (1,187+2,981+26,787) = 30,955 MT/year or 26.7% of the reliability losses. The shut down time of zero output at 23,515 MT/year represents 20.2% of the losses.
- Add the nameplate line with a beta slope of 27 based on benchmarks with similar units and based on the coefficient of variation—beta = 27 is a common starting point for many operations. As improvements occur to reduce variability, the name plate slope beta must be increased to keep the nameplate line to the right of the demonstrated line. Work for a world class beta = 100. The nameplate line in Figure 5 is applied using the menu item for nameplate under the mixture/process reliability icon.
|
Figure 5: Nameplate Line |
|
|
- Now the losses show a significantly different value in Figure 6!
|
Figure 6: Gap Analysis Between Data Points And Both Trend Lines |
|
Total Production = 868,199.1 (Metric Tons) Production Line = Eta 2,775.538, Beta 16.76324 Nameplate Line = Eta 2,880.606, Beta 27 Total Reliability (8.5-100%): Loss = 116,126 (Metric Tons) Process Reliability (%) = 8.5 8.5-11.8%: Loss = 1,187 (Metric Tons) 11.8-61%: Loss = 25,897 (Metric Tons) 61-64%: Loss = 2,981 (Metric Tons) 64-91.8%: Loss = 35,759 (Metric Tons) 91.8-97%: Loss = 26,787 (Metric Tons) 97-100%: Loss = 23,515 (Metric Tons) Efficiency + Utilization (Production - Nameplate): Loss = 48,859 (Metric Tons) Hidden factory losses = 116,126 + 48,859 = 164, 985 MT/year. The hidden factory losses represent 164,985/2775 = 59 days of normal output 70.4% of the losses are between the demonstrated line and the
data points, and |
- Further investigation is required to understand why stair steps exist in Figure 3 and 5. We need to know: are the loss gaps are due to “things that fail” or are due to “grade differences”? Investigate this question by converting the data to Probit-3 format and introducing different symbols for grades by putting each grade into a different dataset.
- Probit-3 data format requires three distinct data entries: 1) the X-axis scale value (in this case daily output), 2) The cumulative quantity occurred in % format (this is Benard’s median rank shown in % format), and the total quantity of data points used. For example the data will have the format 6.6*2.59*10 which would represent 6.6 the X-axis value, 2.59 represents the % Y-axis value, and 10 represents a data set of 10 data points. The Excel spreadsheet mentioned above contains the calculation for the 366 data points used for this analysis. By separating the data into grades of product produced, each data set will maintain it’s proper X-axis value and the proper Y-axis value when the method icon is used and the data format selected by use of the Probit icon for Probit-3 methodology.
- The data for Probit-3 is performed in four steps to avoid confusion and errors.
First, the probability plot position is computed using Bernard’s median rank equation (i-0.3)/(N+04) which is described in C. R. Mische’s paper on A Distribution-Independent Plotting Rule for Ordered Failures, 1979. The “i” value is the plot position number for ranked data going from smallest data value to largest and “N” is the total number of data points in the set.
Second, the data is collected in columns along with the product grade produced.
Third, the collected data is copied and pasted as special values to other columns so it can be sorted accurately.
Fourth, the data is concatenated so it can be imported into WinSMITH Weibull in column form for each grade. - Probit-3 data from the spreadsheet is copied and
pasted into WinSMITH Weibull. Under
the magnifying glass icon, the fit lines are hidden, and under the point
symbol type the quantity is hidden.
The probability plot is shown in Figure 7 with each
symbol highlighting a grade.
|
Figure 7: Probit-3 Plot By Grade |
|
|
- Compare Figure 1 (without grade symbols) to Figure 7 (with grade symbols). Note the “reliability problems” are now identified as assignable to different grades.
- The upper reaches of Figure 7 are shown in Figure 8.
|
Figure 8: Upper Right Hand Corner Data Magnified |
|
|
- Compare Figure 3 (without grade differences) to Figure 8 (with grades identified in the legend). The reasons for the cutback zones are clearly identified as grade dependent with transitions comprised of mixed grades.
- Line slopes for the demonstrated output are assigned in Figure 9.
|
Figure 9: Demonstrated Production Lines By Grade |
|
|
- Figure 9 says grade B products are produced at
2632/2776 = 94.8% of the rate of grade B products. Grade C products are produced at
2373/2776 = 85.5%
of the rate of grade A products.
Grade B and grade C products run through the process at slower
production rates and the cutback zones are not properly identified as
reliability problems (only the down time and the transition zones). The redefinition of the hidden factory
is shown in Figure 10.
|
Figure 10: Gap Analysis Between Data Points And Both Trend Lines Restated |
|
Total Production = 868,199.1 (Metric Tons) Production Line = Eta 2,775.538, Beta 16.76324 Nameplate Line = Eta 2,880.606, Beta 27 Total Reliability (8.5-100%): Loss = 116,126 (Metric Tons) Process Reliability (%) = 8.5 8.5-11.8%: Loss = 1,187 (Metric Tons) 11.8-61%: Loss = 25,897 (Metric Tons) ßTransition zone, a hidden factory 61-64%: Loss = 2,981 (Metric Tons) 64-91.8%: Loss = 35,759 (Metric Tons) ßTransition zone, a hidden factory 91.8-97%: Loss = 26,787 (Metric Tons) 97-100%: Loss = 23,515 (Metric Tons) ßDowntime Removing the grade production
rates shows the Reliability Losses are:
Efficiency + Utilization (Production - Nameplate): Loss = 48,859 (Metric Tons) Hidden factory losses = 54,470 + 48,859 = 103,329 MT/year. The hidden factory losses represent 103,329/2775 = 37 days of normal output Based on the recognition of slower grades in production the losses are restated: 52.7% of the losses are between the demonstrated line and the
data points, and |
- Figure 10 shows the losses are about the same
magnitude between reliability issues and efficiency/utilization issues
with almost half of the reliability losses associated with downtime
issues. Putting all the losses on
one graph is shown in Figure 11.
The graph was captured as a soft copy in WinSMITH Weibull and
pasted as a bitmap file into Powerpoint. The colored zones were drawn by turning
off the grid under the drawing features.
Also using the drawing features, the Autoshape/Lines/Freeform
feature was used. Simply click on
where the freeform should start and remove your finger from the mouse and
pull the mouse pointed to where the line ends and click again with
repetition of the process to close the freeform. The colored zones of the freeform were
formatted for a transparency value of 50% so you can see through the
colored zone to find the symbols.
|
Figure 11: All Losses Shown On
One Sheet With Grade Considerations |
|
|
- Of course reducing downtime for the process to reduce reliability losses is an important no-brainer issue. Increasing the predictability of process output by eliminating problems restricting output will increase slope of the trend lines. Variability in output reduction is an important issue to reduce losses and make the payroll of the plant more consistent. The low reliability is driven by the small amount of product A produced.
- Perhaps technology improvements can be implemented to reduce the apparent losses from grade changes.
- None of these problems cure themselves. You’ve got to make a change to get a change! Fix the problems and generate wealth for the corporation!
Want More Process Reliability Information?-
· Production Reliability Example With Nameplate Ratings
· Key Performance Indicators From Weibull Production Plots
· Process Reliability Plots With
Flat Line Slopes
· Process Reliability Line Segments
· Automating Monthly Weibull Production Plots From Excel Spreadsheets
·
Papers
On Process Reliability As PDF Files For No-charge Downloads
- New
Reliability Tool for the Millennium: Weibull Analysis of Production Data
- Process
Reliability and Six-Sigma
- Process
Reliability Concepts
Download a copy of this problem of the month involving process reliability plots with probit analysis by clicking here.
Return to
Barringer & Associates, Inc. homepage
Last revised 05/02/2005