Downhole Sensor Reliability Index (DSRI)

GMI-OG-020Oil and GasInfrastructure-Dependent

Definition

Downhole Sensor Reliability Index, or DSRI, is an estimated probability that anomalous sensor readings fed to AI production optimization and reservoir models reflect genuine asset and process conditions rather than sensor degradation.

The Measurement Gap

Vihaan is a production engineer responsible for about ninety ESP-lifted wells on a heavy-oil asset in the San Joaquin Valley. Each well has a downhole sensor package (motor temperature, motor current, intake pressure, intake temperature, discharge pressure, vibration) feeding a production optimization platform that adjusts pump speed in real time against operating envelopes. The platform has been a quiet success for two years. Pump runtimes have improved, intervention frequency is down, and the optimizer''s decisions on pump speed and gas separation have produced incremental oil at a cost the asset manager can defend.

The incident came on a well that had been running normally for fourteen months since its last workover. At 02:14 on a Saturday morning, the optimizer received a motor temperature reading from the downhole sensor that was eighteen degrees above the well''s trending baseline. The platform''s logic correctly identified the reading as elevated and increased the well''s pump speed slightly, which had the effect of cooling the motor: the fluid throughput rose, more heat was carried away, and the next reading was back closer to trend. The optimizer logged the response as a successful temperature management action. Over the next six hours, the same pattern repeated four more times. Each time the optimizer''s response brought the temperature back down. Each adjustment was inside the pump''s allowed envelope. The platform''s dashboard showed a well being managed competently by the AI.

What the optimizer did not realize was that the motor temperature sensor''s thermistor had started to fail. The drift was real on the sensor, not on the motor. The actual motor temperature was within trend the whole time. The optimizer had been responding to false elevations by pulling more fluid through a motor that did not need more cooling, which had the side effect of operating the pump above its optimum point. By the time Vihaan came in Monday morning and looked at the well (flagged automatically because the cumulative pump-speed adjustment exceeded a daily threshold) the pump had been operating outside its sweet spot for thirty-six hours. The well''s production was off, and the pump''s motor current was showing wear signatures that suggested the bearings had taken stress. Two weeks later the pump tripped, and the workover that followed found the motor in the early stages of damage from operating off the curve.

The post-incident analysis was clear. The motor temperature reading was a single number from a single sensor. The optimizer''s logic treated the number as the motor''s actual temperature without independently checking whether the reading was consistent with the other sensors on the well. Motor current was within trend, indicating the motor was not actually working harder. Intake pressure was within trend, indicating the reservoir was not changing. The discharge pressure was within trend, indicating the fluid pathway was healthy. Every cross-check sensor said the motor temperature should be normal. Only the motor temperature sensor itself said otherwise. The optimizer had no logic to weigh the cross-check against the primary reading. It trusted the reading.

Legacy ESP monitoring metrics do not catch this. Sensor uptime was high: the thermistor was producing readings, just wrong ones. Optimizer response time was excellent: every reading was being acted on within minutes. The well''s production was being managed inside its envelope. The drift was a sensor degradation, not a sensor failure, and the standard "sensor offline" alerts did not fire because the sensor was producing readings. The standard "outside envelope" alerts did not fire because the optimizer''s responses kept the well inside its envelope.

What is missing is a metric that asks whether the AI''s incoming sensor data is actually consistent with the physical reality of the well or whether the optimizer is acting on a degraded measurement.

Formula

Downhole Sensor Reliability Index, or DSRI, is an estimated probability that anomalous sensor readings fed to AI production optimization and reservoir models reflect genuine asset and process conditions rather than sensor degradation. The score is computed per sensor through multi-sensor consistency checks and drift trend analysis, then aggregated to a well-level number. The KPI exists to answer one decision: when the AI sees a reading that triggers action, is the reading describing reality or is it describing a failing sensor, and is the AI making that distinction at all?

The structural form is a per-sensor health probability that combines two ideas. The first is cross-correlation consistency. Sensors on the same well are physically correlated: motor temperature should track with motor current, intake pressure should track with production rate, discharge pressure should track with the fluid pathway state. When a sensor''s reading diverges from what the correlated sensors predict, the divergence is evidence that the sensor itself has drifted. The second is drift trend analysis. A sensor whose readings are slowly walking away from the predictions of the correlated sensors over weeks is showing the signature of degradation; a sensor whose readings track the correlated sensors is healthy. DSRI per sensor is one minus the magnitude of the divergence between the sensor reading and the correlated-sensors prediction, normalized to the sensor''s acceptable tolerance band. A perfectly consistent sensor has DSRI of 1.0. A sensor that has diverged a full tolerance band has DSRI near zero. The well-level DSRI is the average across the sensors on the well.

The correlation model is the methodological core. The model takes the form of an empirical regression or, in more sophisticated implementations, a physics-informed model that predicts each sensor''s reading from the others under expected operating conditions. The model is built from a sufficient window of healthy historical data (typically thirty to ninety days of operation on the well) and is refit when the well''s operating regime changes substantially. The model''s predictions are the baseline against which the live sensor readings are evaluated, and the quality of the model is the limit on how cleanly DSRI can detect drift.

Proposed formula, pending validation pilots.

At a Glance

Category
Subsurface Decision Quality
← Regulatory Activity Scheduling Segregation Index (RASSI)Drilling Optimization Deferred Quality Index (DODQI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment