<\/span><\/h2>\nAccuracy<\/strong><\/p>\nAccuracy is the closeness of measurement to the true process value. For machine learning:<\/p>\n
\n- Process pH: accuracy of \u00b10.02 pH or better, with automatic temperature compensation.<\/li>\n
- Dissolved oxygen: accuracy of \u00b10.1 mg\/L or 1% of reading, whichever is greater.<\/li>\n
- Conductivity: accuracy of \u00b11% of reading, with documented cell constant traceability.<\/li>\n
- Turbidity: accuracy of \u00b12% of reading or \u00b10.02 NTU below 1 NTU.<\/li>\n<\/ul>\n
Completeness<\/strong><\/p>\nCompleteness is the absence of gaps. Models typically tolerate less than 0.5% missing samples per week before predictions degrade. Instruments must therefore include:<\/p>\n
\n- Self-diagnostic status registers so gaps can be flagged, not silently filled.<\/li>\n
- Local logging so a network outage does not destroy evidence.<\/li>\n
- Documented recovery behaviour after a power interruption.<\/li>\n<\/ul>\n
Timeliness<\/strong><\/p>\nTimeliness is the age of the reading when it reaches the model. For real-time control:<\/p>\n
\n- Analyzer response time under 30 seconds for wetted electrodes and under 2 minutes for wet chemistry.<\/li>\n
- Modbus polling latency under 500 ms in the plant network.<\/li>\n
- Timestamps at the transmitter, not at the ingestion server, to avoid ambiguity during network jitter.<\/li>\n<\/ul>\n
Consistency<\/strong><\/p>\nConsistency means identical process states produce identical readings across time and across sensor units. Requirements include:<\/p>\n
\n- Drift under 2% of range per calibration interval.<\/li>\n
- Standardized units and register layout across every sensor of the same family.<\/li>\n
- Reference chemistry traceability that permits cross-plant comparison.<\/li>\n<\/ul>\n
<\/span>What Happens When Data Quality Fails<\/span><\/h2>\nWater utilities that have published post-mortems on failed AI pilots report broadly consistent failure modes:<\/p>\n
\n- Silent drift:<\/strong> an aeration dissolved oxygen sensor drifts by 0.4 mg\/L over 60 days without alarm; the model treats the drift as a genuine trend and slowly detunes aeration.<\/li>\n
- Communication gaps:<\/strong> an ammonia nitrogen sensor drops out for 6% of samples during monsoon-season network instability; the model’s predictions on nitrification degrade until the sensor is repaired.<\/li>\n
- Time drift:<\/strong> transmitters and ingestion servers use different clocks; the model attributes cause and effect incorrectly and issues counter-productive dosing recommendations.<\/li>\n<\/ul>\n
Each of these failure modes is preventable at the instrument level, and each is expensive to detect after the fact.<\/p>\n