F
Member
【Editorial Discussion】
Historian exception and compression settings are often tightened “so we do not miss anything,” then the disk fills, and a short transient is still missing from the trend. Those two outcomes are compatible: a tight deadband on noisy tags writes almost every sample, while a filter or a too-wide band on a fast tag can still skip the peak you wanted.
1) Apply different policies to noisy analog tags, to status/digital tags, and to fast transients you have promised to capture. A single plant-wide deadband is usually wrong for all three. If a tag is used for protection or for a known spike, its collection must be specified against that use, not against average storage.
2) Know whether you are saving exceptions, compressed segments, or both, and what the retrieval will reconstruct. A trend that looks triangular after compression may have hidden the flat top of a trip. Retrieve with the same settings you will use in an investigation, not only with a raw dump on a test tag.
3) Watch volume by tag. A handful of noisy inputs can dominate storage. Fix those first — shielding, filtering at the source, or a more appropriate band — before buying another disk or loosening every tag.
4) Test with a known pulse of defined width and amplitude. If that pulse does not survive into the historian under the production settings, the settings do not meet the requirement, no matter how full the disk is.
Open question: have you had a storage crisis and a missed transient on the same system? How do you now decide a deadband — from process knowledge, from noise measurements, or from a default? Please share what you changed.
Historian exception and compression settings are often tightened “so we do not miss anything,” then the disk fills, and a short transient is still missing from the trend. Those two outcomes are compatible: a tight deadband on noisy tags writes almost every sample, while a filter or a too-wide band on a fast tag can still skip the peak you wanted.
1) Apply different policies to noisy analog tags, to status/digital tags, and to fast transients you have promised to capture. A single plant-wide deadband is usually wrong for all three. If a tag is used for protection or for a known spike, its collection must be specified against that use, not against average storage.
2) Know whether you are saving exceptions, compressed segments, or both, and what the retrieval will reconstruct. A trend that looks triangular after compression may have hidden the flat top of a trip. Retrieve with the same settings you will use in an investigation, not only with a raw dump on a test tag.
3) Watch volume by tag. A handful of noisy inputs can dominate storage. Fix those first — shielding, filtering at the source, or a more appropriate band — before buying another disk or loosening every tag.
4) Test with a known pulse of defined width and amplitude. If that pulse does not survive into the historian under the production settings, the settings do not meet the requirement, no matter how full the disk is.
Open question: have you had a storage crisis and a missed transient on the same system? How do you now decide a deadband — from process knowledge, from noise measurements, or from a default? Please share what you changed.