Data infrastructure construction - メーカー・企業と業務用製品 | イプロスものづくり

[Case Study] 70% Reduction in Sudden Failures | Establishment of Predictive Maintenance Data Infrastructure

"Although there are sensors, they cannot be used" and "PoC stops." To the on-site personnel: We are publishing the entire procedure for the implementation that reduced unexpected failures by about 70% in 4 steps.

"Although there is sensor data, it cannot be used on-site." "We tried a PoC, but the accuracy was insufficient, and we couldn't get management approval." - Predictive maintenance in the manufacturing industry often stalls at this "one step away." This case study reveals three challenges that hinder predictive maintenance in environments troubled by unexpected machine stoppages: 1. Opportunity loss due to unexpected stoppages during busy periods, 2. Aging mechanics and the personalization of diagnostic know-how, 3. The barrier of lacking a data infrastructure that halts PoC efforts. We will share the steps taken to overcome these challenges in four stages (sensor maintenance → data integration → assetizing "intuition" → operational deployment). We reduced the number of unexpected failures by about 70%, decreased the data formatting and preprocessing workload by about 60%, and detected anomalies an average of three days in advance. This is a record of our support in establishing "AI that is actually used on-site and continues to function," rather than just stopping at a PoC. NTP's approach is to leave behind "moving assets" that continue to operate in the field, rather than "thick reports." Recommended for those who: - Want to implement predictive maintenance but are struggling with data infrastructure development. - Have tried a PoC but did not achieve sufficient accuracy and have not received management approval. - Face challenges in knowledge transfer due to the retirement of veteran mechanics. - Have sensor data but are not fully utilizing it.