The Nature of Data in Enterprise Software.
In my first article of this series, I had talked about the reasons that interested me in this topic .
If I am going to design an enterprise-wide package for a steel producing company, I must understand and consider the nature of data: where it is generated, by whom it is entered, by whom it is used and for what purpose.
Accountants and production engineers in a Steel Plant typically look at data in different ways.An accountant needs consistent and verifiable data, either authenticated by the generating department, or generated by the Accounts department itself.The data is posted :either into Ledgers and Journals, or into a Computer Program by office staff trained and held accountable for accurate clerical work.Books of Account, and Management Information Reports for presentation to the Corporate Office or Statutory Authorities are prepared using this data.
Production shops, on the other hand, represent important "generating departments" in Steel Plants.The generating method is the time-honored production report.In a typical production shop (assuming that there is anything typical in a collection of production shops using such a wide variety of inputs and producing an equally wide variety of outputs) the report generation stage proceeds something like this:
Coal or iron ore stored in silos, a ladle of molten steel, an ingot, a bundle of rods bearing a specific batch number, a coil, a packet of sheets- all these can be inputs, and most of them can be outputs.It depends on the particular shop we are considering.
The input details are entered in the production report form.The unique identification no. has to be verified by the production department( for solid steel this number is usually written on the steel surface, else embossed in steel tags attached to the packet bundling wire, else pasted as paper labels on the top of the coil).Production practice may or may not call for weighing of the batch before taking it into production: in case the batch is not weighed, the output weight is taken from the production record of its previous process.
Output is usually weighed at the end of the process.On occasion weigh scales are not conveniently located, or may be under breakdown. In that case, typically "calculated" weights are used, with verification of output weights being (or not being) done at a different time and location.Sometimes the verification takes place weeks or months after production at the start of the next process.The delay is sometimes the result of a decision by Production Planning Department; sometimes the constraint is a purely physical one( material lying at the bottom of a heap).The environment is typically hot and dusty; the data is entered round the clock by production workers and supervisors primarily charged with monitoring and maintaining the production process.
In this environment the generated data is bound to be imprecise.The amount of imprecision in production data would depend entirely on the management structure, and also the data field involved.
Why should any attention be paid to imprecise data? Because they are ubiquitous, wide-ranging ,unbiased , consistently produced over long periods of time and cost virtually nothing to generate.Production engineers use statistical methods to analyze these reports to get valuable information which form the basis of quality improvement and cost-reduction initiatives.This information is valuable, available nowhere el else, and should be considered to be a data warehouse to be profitably mined at a future date.
Once we acknowledge the nature of production data, we can look at using a subset of this data, properly filter it to ensure consistency, and port it directly to the ERP. The filtration process will call for manual intervention; the more precise the shop-floor data, the less the filtration effort required.The architecture for the incorporation of production data into ERP software that emerges is a Data Warehouse application layer located below the ERP , to be used for Data Entry and Verification as well as Data Mining.
The only problem that remains is the impatience of Senior Management with the non-intuitive science of Statistics.If anyone tries to micro-manage a production process using discrete bits of data rather than taking a statistical view, he could end up by over-correcting the process, creating new problems rather than solving old ones.It is better to leave production management to professionals and hold them accountable for results.
I shall discuss my experience with the demo Data Mining program I installed in my next article.
No comments:
Post a Comment