This is the fifth essay in a series expanding on an article I wrote for Sensors titled "A Twelve-Step Sensor Selection Checklist." This month, I'll look at the phase of the sensor selection process in which you review your error analysis. In this step, you decide if a particular sensor meets your measurement requirements. Some people balk at doing detailed analysis, saying it's unnecessary. Others will want to just try a sensor and see how it works. In most cases, the analysis approach beats the cut-and-try method hands down. You've come this far. Don't waste the effort.
In this analysis, you estimate the likely errors encountered with a sensor. Essentially, you answer the question: How close will this choice come to meeting my measurement requirements? Believe me, the analysis is worth doing and documenting. Someone always questions your choice. So have your answer ready.
The review can be difficult and lengthy and may require some serious modeling—using Finite Element methods or other analytical techniques—to get to a realistic appraisal of likely results. Remember, however you arrive at your estimate, realize that it is but one component in the calculation of measurement uncertainty. In the overall process, you must include all the components in the total measurement uncertainly calculation, combining calibration uncertainties and those due to influencing factors such as varying ambient temperature and other conditions.
Often the calculation is a simple root sum of the squares of the uncertainty components. You have to decide on the confidence level appropriate for your organization or the outcome of your process.
The ISO standards on measurement quality provide good guidelines for making these determinations. You can find these on the measurement uncertainty pages of the National Institute of Standards & Technology's (NIST) Web site. NIST is clear about understanding, evaluating, and combining measurement uncertainties and gives several examples of the calculation procedure.
If you decide a sensor won't meet your needs, you either try another sensor or relax one or more of the measurement requirements. The requirements may be unrealistic, given the state of the art or the cost of different measurement devices, and there is almost always a tradeoff.
Other possibilities are:
- Do you really need one measurement device model to cover the full and expected range of conditions?
- Can you limit the measurement span?
- Can you meet the requirement with a pair of sensors or several different sensors, each covering different parts of the overall range?
- Is the measurement requirement less or the uncertainty easier to realize in a portion of the target measurement span?
There are lots of angles to pursue. However, there are times when no standard sensor will do the job. At that point, you have to admit you are chasing real but unattainable goals. But even at this point, you are not out of options.
You can explore the possibility of modifying an existing sensor model. Most vendors do "special" regularly. Another option is to sponsor a totally new design, one aimed at meeting your needs. Some companies and government agencies have sponsored full-scale development efforts to expand the state of the art of the measurement technology.
Another option is to use the best available sensor to merely monitor the process. If it is a process where you have a substantial knowledge of the other influencing parameters, it's possible to measure and control them better, rather than the desired variable.
For example, a few years back, one of the difficult challenges in the steel industry was to control the process of galvannealing zinc-coated steel strips. The process was very valuable and difficult to control because of the many process parameters involved.
The temperature of the strip was a key variable, and it was nearly impossible to measure because typical strip speeds were hundreds of feet per minute. However, a few steel companies were able to meet coating specifications by careful control of all the input variables and empirically "walking" their way into the correct process window and staying there for long periods of time.
When at last a suitable temperature sensor was developed, those companies that had gone the extra mile to get the influencing variables under tight control gained extra process line productivity and yield by adjusting some of the other variables, such as time and temperature post coating, while increasing line speed.
I hope this helps. I'd appreciate feedback from those who have been through similar experiences. In any event, I'll move on to the sixth step in the next article.