The operation

A Norwegian marine ingredient manufacturer processing raw marine biomass into oil and meal products for aquaculture and human nutrition markets. Highly instrumented processing lines with established quality control procedures. Existing monitoring focused on process compliance and quality specification adherence.

The specific challenges of marine ingredient processing

Marine ingredient processing operates with variable raw material quality. Raw material composition changes seasonally and by catch location. The processing parameters that produce optimal yield and quality are not fixed; they depend on the characteristics of the incoming raw material in ways that experienced operators understand but that are difficult to encode in static process settings.

The result is that the 'normal' for this operation is genuinely variable. Static thresholds struggle to distinguish a parameter variation that is appropriate given the current raw material profile from one that represents a genuine process deviation. False positives are common. Early signals of genuine deviations are sometimes missed.

Why dynamic baselines matter here

Eyer’s three-dynamic-baseline approach is particularly well-suited to processing operations with variable inputs. The baselines are learned from historical behaviour, capturing the normal range of variation historically associated with good product quality and yield, and naturally incorporate the contextual variation that makes static thresholds inappropriate.

The baseline for a processing parameter under a specific raw material profile is different from the baseline under a different profile. Eyer learns this from the historical data, without requiring manual configuration of the contextual rules.

What the analysis found

The Fast Forward analysis identified anomaly patterns in the processing line data that correlated with downstream quality and yield variation. Several of these patterns were detectable in the processing data hours before their impact was visible in quality measurements.

The correlation structure was informative: the primary deviation signals were upstream of the quality measurement points, in processing parameters that the existing system monitored but did not analyse for fingerprint deviations. The correlation model that Eyer built from the historical data made the causal relationship between upstream processing behaviour and downstream quality outcomes explicit.

The operational knowledge dimension

In a processing operation where the knowledge of how to respond to variable raw material quality is largely carried by experienced operators, the operational fingerprint has a second value: it encodes the relationship between processing parameter behaviour and product outcome in the data. New operators have access to the historical pattern. The knowledge does not depend entirely on the experience of the individuals currently running the line.