Data Sampling in a Real Time World – by Don Zerrip CxGBS®
Apr 5, 2012 CxGBS Articles
Data used to be a very expensive commodity. Not too long ago scientific sampling methods most likely used an hourglass or egg timer device, pencil, and paper. Periodic samples could be taken as the last grains of sand passed through hourglass. One or two measurements could be documented and the time period was reset by turning over the hourglass. With modern low cost technology the hourglass has been replaced with a computer sampling large amounts of data a (nearly) the same time and recorded into memory. Time stamped data can provide an adequate representation of the process under study at a reasonable cost if the proper sampling plan is used.
Today’s methods for building automating system (BAS) sampling are typically periodic fixed time intervals or change of variable (COV) sampling. Both have their usefulness. By sampling multiple data at a synchronized fixed period interval, many variables can be measured at the same time and relationships between commands, actions, and reactions can typically be determined. Usually a tradeoff must be made between sampling rate, speed at which sampled data can change, and memory capacity. If you are measuring room temperature, barometric pressure, or outdoor air temperatures then sampling every 15 minutes usually provides adequate results. Sampling every minute would provide you with a better representation of the process, but if you are taking twelve measurements on a hundred variable air volume boxes in a building every minute the computing power and data load becomes immense with only a marginal increase in process understanding. Slow processes do not require high sample rates but you must remember that a failure can be masked within a slow sampling rate.
On the other hand taking COV measurements only requires knowing what the present value is and if it has changed since the last measurement. Sampling rates can be very high but the value is only recorded when the value changes. This is great for a value that does not change often. Time resolution can be near instantaneous and memory requirements are small. The downside is if the value changes frequently a plethora of data can be produced in a very short time period. Typically the resolution of the measurement needs to be considered unless you want to know exactly when and how often the temperature in a room changed by 0.01 degree Fahrenheit.
Both methods of monitoring have their place. Data requirements for sampling should consider the process being monitored and cost of the data. A frequent sampling of large amounts of data does little good unless continuous monitoring methods are in place. Knowing how equipment performs is a powerful tool that needs careful consideration. Sampling plans are an important part to continuous monitoring of system performance to provide process insight at a reasonable cost and prevent data overload.
Tags: BAS Sampling, COV Sampling, Data Sampling
