On the Topic of Programming 2

2 09 2012

September 2, 2012

(Back to the Previous Post in the Series) (Forward to the Next Post in the Series)

Over the course of the last six months I developed several efficiency reports for the machining facility where I work, measuring the average time required to complete a customer’s part compared to the engineering standard “target” run time for producing the part.  Sounds simple, right?  If the engineering target is 30 widgets per hour, and only 20 widgets per hour are produced, then the average efficiency is 20/30 = 66.67%.  It does not take a person with a degree in mathematics to arrive at the conclusion that the widgets are not being produced as quickly as expected.  What is the scope of the measurement time frame for the average: a year, a month, a week, a day, one work shift in a day, a single machining center for a day, a single machining center and employee, or something else?

OK, now flip the engineering standard run times a bit so that the times are a little more consistent with the machining facility where I work.  It might take four hours, six hours, 10 hours, 20 hours, or possibly even 200 hours to complete a single operation at a machining center (one of several operations in the manufacturing process for the widget) to produce a single widget.  With the long run times of the operations, calculating the efficiency of a machining center or an employee for a specified period of time was a daunting task… a task that I was asked to solve roughly 12 years ago (with the help of Oracle Database 8.0.5).

The hours per widget run time (rather than widgets per hour) situation presented various problems for calculating employee efficiency, especially when an operation at a machining center did not complete before the end of an employee’s shift.  Consider a situation where an operation is expected to require eight hours to complete, but only when the tooling used at the machining center is new.  As such, the first shift employee installs new tooling in the machining center every morning before starting a new widget.  The first shift employee spends about 30 minutes finishing up the widget that was started the previous night, changes the tooling, and then starts the machining on the next widget.  So, the first shift employee reports that one widget (started the previous night) completed after 0.5 hours and the second widget completed in 7.5 hours.  The first shift employee’s efficiency, considering that two widgets were completed during his shift, is easily calculated as (8 + 8) / (0.5 + 7.5) * 100 = 200%.  The second shift employee’s efficiency is a consistent 0% because the operation for the widget never completes during his shift because the sharpness of the tooling deteriorates through the day (thus causing the machining operation to take longer).  This obviously leads to odd questions: 1) Why is the second shift employee more efficient when the first shift employee is on vacation (equivalent to the question in Oracle performance tuning: why does my report run faster when it rains Monday mornings?)?  Why is the second shift employee more efficient when working a nine or 10 hour shift, rather than an eight hour shift?  The fun questions that one is able to answer when familiar with the data…

When an employee starts working on a machining operation, a labor ticket transaction is created detailing the fact that the work for the machining operation is in-process.  When the machining operation completes (or when it is time to leave for the day), the employee closes the labor ticket transaction and reports the number of widgets that completed.  These labor ticket transactions are conveniently stored in a table named LABOR_TICKET.  Back in 2006 or 2007, I decided to take another attempt at solving the efficiency problem – after all, I now had access to analytic functions in Oracle Database (analytic functions did not exist in Oracle Database 8.0.5).  I expected that by extending the time range beyond a single employee and/or shift, I might be able to extract useful efficiency information from the database.  I put together a rather long SQL statement that looked something like this:

SELECT DISTINCT
  1 TYPE,
  WO.PART_ID,
  TO_NUMBER(LT.WORKORDER_SUB_ID) WORKORDER_SUB_ID,
  LT.OPERATION_SEQ_NO,
  TRUNC(SYSDATE-30) SHIFT_DATE,
  LT.EMPLOYEE_ID,
  LT.RESOURCE_ID,
  SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID) HOURS_WORKED,
  SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID) GOOD_QTY,
  NULL HRS_PC,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO)),2) AVG_HRS_PC_TIME_ALL,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID)),2) AVG_HRS_PC_TIME_EMP,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID)),2) AVG_HRS_PC_TIME_RES,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID)),2) AVG_HRS_PC_TIME_EMP_RES
FROM
  WORK_ORDER WO,
  LABOR_TICKET LT
WHERE
  WO.TYPE='W'
  AND WO.SUB_ID='0'
  AND WO.PART_ID='WIDGET1'
  AND LT.SHIFT_DATE BETWEEN TRUNC(SYSDATE-30) AND TRUNC(SYSDATE)
  AND WO.TYPE=LT.WORKORDER_TYPE
  AND WO.BASE_ID=LT.WORKORDER_BASE_ID
  AND WO.LOT_ID=LT.WORKORDER_LOT_ID
  AND WO.SPLIT_ID=LT.WORKORDER_SPLIT_ID
UNION ALL
SELECT DISTINCT
  2 TYPE,
  WO.PART_ID,
  TO_NUMBER(LT.WORKORDER_SUB_ID) WORKORDER_SUB_ID,
  LT.OPERATION_SEQ_NO,
  TRUNC(SYSDATE-60) SHIFT_DATE,
  LT.EMPLOYEE_ID,
  LT.RESOURCE_ID,
  SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID) HOURS_WORKED,
  SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID) GOOD_QTY,
  NULL HRS_PC,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO)),2) AVG_HRS_PC_TIME_ALL,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID)),2) AVG_HRS_PC_TIME_EMP,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,RESOURCE_ID)),2) AVG_HRS_PC_TIME_RES,
  ROUND(DECODE(SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID),0,0,
    SUM(LT.HOURS_WORKED) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID)/
      SUM(LT.GOOD_QTY) OVER (PARTITION BY WO.PART_ID,LT.WORKORDER_SUB_ID,LT.OPERATION_SEQ_NO,EMPLOYEE_ID,RESOURCE_ID)),2) AVG_HRS_PC_TIME_EMP_RES
FROM
  WORK_ORDER WO,
  LABOR_TICKET LT
WHERE
  WO.TYPE='W'
  AND WO.SUB_ID='0'
  AND WO.PART_ID='WIDGET1'
  AND LT.SHIFT_DATE BETWEEN TRUNC(SYSDATE-60) AND TRUNC(SYSDATE-31)
  AND WO.TYPE=LT.WORKORDER_TYPE
  AND WO.BASE_ID=LT.WORKORDER_BASE_ID
  AND WO.LOT_ID=LT.WORKORDER_LOT_ID
  AND WO.SPLIT_ID=LT.WORKORDER_SPLIT_ID
...

The above SQL statement grew in length substantially based on the number of time periods that I selected to compare.  This data was returned to a program for display purposes, so reformatting the output was always a possibility.  Consider a situation where the programmer is unfamiliar with Oracle specific SQL and is asked to generate the same output in a program that he is constructing.  For a program seeking to measure efficiency, the programmer’s solution would likely involve repeated executions of SQL statements to retrieve the bits and pieces of information that need to be presented – this row by row processing will likely be very slow over a high latency WAN (VPN) connection (but certainly faster than never receiving the information, so the programmer will still be a hero), and could very well hinder the performance of the Oracle Database server.

The above solution is good, but problems such as the progressive wearing of the tooling cutter, as mentioned earlier, could still lead to unexpected differences in efficiency of different employees that are working as fast as is possible.  Fast forward a few years.  The right questions are asked from people with a more complete understanding of efficiency measurements – examine the efficiency calculation problem from an entirely different angle.  Not long ago I was handed an example of a simple efficiency report, and asked to reproduce that report with live data, output in a Microsoft Excel spreadsheet.  The example was little more than a simple sketch, so for this blog article I put together a color-coded example of the report format in Microsoft Excel:

A little explanation is required.  The entirely different angle for calculating employee efficiency in the execution of widget operations involves not looking at the date of the labor ticket transaction, or even the number of widgets produced by a particular employee in a particular time frame.  Instead, the date of the manufacturing batch’s (lot in this particular ERP system) completion date, the quantity produced in the batch, and the total hours to execute a single machining operation for the lot become the criteria when comparing against the engineering standards to determine efficiency.  The manufacturing batch’s completion date is used to divide the batches into specific timeframes (in this case weeks: Monday through Sunday).  All employees working on a particular operation, where the batch completion date is in a certain week, will receive the same efficiency rating for that particular operation (EFF_WEEK1, EFF_WEEK2, EFF_WEEK3, etc.) as all other employees working on the same operation with the same batch completion date range.  It is not uncommon for employees to generate labor transactions for multiple operations for production of the same part, as well as operations for different parts that have batch completion dates in the same week, so the employee’s efficiency rating (W_AVG_EFF1, W_AVG_EFF2, W_AVG_EFF3, etc.) weights the individual part efficiencies based on the percentage of machining time the employee spent in a given operation compared to all of the other operations the employee worked.  The employee’s efficiency rating (Avg W_VE_EFF) is the simple average of the employee’s weekly efficiency ratings.

In the above example report, the blue colored text is a static label.  The red colored text is a calculated static label that shows the Monday through Sunday date range for the week.  The black colored text is a simple “rip and read” from the database – no calculation is required.  The purple colored text indicates a calculated value.  The numbers at the right will not appear on the completed report, but are included to indicate which EFF_WEEKn and AVG_WEEKn values will be identical if the employee worked on an operation whose batch closed in the given week (notice that there are blanks in columns of the sample report, indicating that the employee did not work on that operation in the batch completion week).

The programming challenge is to determine the number of SQL statements that would be needed to retrieve the information from the database, and the number of times those SQL statements would need to be executed.  Would the number of SQL statement executions depend on the number of employees?  Would the number of SQL statement executions depend on the number of different part operations whose batches closed in a particular week?  Would the number of SQL statement executions depend on the number weeks included in the report?  Before you think about the problem, we should probably investigate the data organization in the various tables.  The picture below shows the tables (and the specific columns) required for the report, showing the tables for the engineering standards in green boxes and the tables needed for the labor transaction analysis in black boxes.  The lines show the association of the data in the different tables.

(The above image was adjusted 7 hours after this blog article was initially posted.  The linking between the green OPERATION table and the green WORK_ORDER table was corrected, and the red colored text was added to indicate whether the table was to be used for retrieving the engineering master standards (WORKORDER_TYPE=’M’) or information from the production work order batches (WORKORDER_TYPE=’W’ or TYPE=’W’).)

It might also be interesting to think about what processing will be performed on the data returned by the SQL statements – that could be a pivotal design decision for the SQL statements.  Curve balls are to be expected – what is currently an examination of four weeks’ worth of efficiency numbers today might be an examination of 52 weeks tomorrow.  Oh, here is a thought, what if instead on the next day it is important to focus on the various part efficiencies and the employees who worked on the operations, rather than focusing on the employees and the various parts that the employees worked on?

Part 3 of this series will attempt to answer some of the above questions.

Late addition, sample of a completed report in Excel:


Actions

Information

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s




Follow

Get every new post delivered to your Inbox.

Join 139 other followers

%d bloggers like this: