Quantify

From Engineer-it
Jump to navigation Jump to search
References
Internal links

Engineer-it

Strategies

Examples

Structural analysis [link to a page on Analysis Modelling in the Structural Enginnering Chapter - not yet available]

In an engineered process, It is important to quantify those features that can be quantified.

Predictive Modelling

Predictive modelling is the use of mathematical representations to estimate the behaviour of systems. Such modelling can be very successful for physical system, e.g. engineering mechanics, but can be less effective when dealing with systems that depend on human behaviour e.g. economic modelling.

We tend to think of modelling as a determinate process - that has a correct answer. That is how it is taught in education where learners are normally presented with a model and are required to produce results that are either correct or wrong. In real world modelling, the overall process is non-determinate and has to be addressed using the explicit top-down strategy.  Here is how that strategy applies to predictive modelling:

Inception

Gather information about the context such as the geometry, material, external action, etc. Establish a set of requirements for the model in terms of features that need to be modelled,expected accuracy, etc.

Conception

Identify the types of model that may be used and assess their validity. Carry out a validation analysis. The validation question is: ‘Is the model capable of satisfying the requirements?’  Two ways of validating a model are:
  • Assumptions assessment: Compare each assumption made for the model with the expected behaviour of the system.  Make a judgement about validity on the basis of this information.
  • Physical testing: Assess information from commmissioned tests or from published tests.
On the basis of the results of the validation analysis and other information, choose the model that is most suitable in the context.

Production

Set up the data and run the model. Verify the results i.e. seek answers to the question ‘Has the model been correctly implemented?’ Assume that there may be errors. In safety critical contexts, work to a results acceptance process before using the results.

At all stages it is important to operate in a critical thinking mode where information is regularly challenged and tested for reliability. One must keep in mind that the results from predictive models are always approximations to real behaviour.

Data

“In God we trust. Others must bring data.” anon.  Using data instead of guessing is obviously preferable but:

“Lies, damned lies and statistics”  anon. It is essential to be sceptical about all data and the results that emerge from its use.

For example, use of data is a core strategy in the regulation of drugs but it is not easy to ensure that the processes and the conclusions are reliable. Bias in data and in its interpretation easily arises - both accidentally and deliberately.