BD-CUBE
1.1 Overview of BD-CUBE
- Hitachi High-Tech Solutions Corp., Japan
Hitachi BD-CUBE is an analytical diagnosis system which applies mathematical modeling technique, using big data of field control system. BD-CUBE is aimed at stable operation of the plant. It utilizes plant information to improve “quality and productivity”. In addition, it detects “equipment failure signs” and provides feedback.

1.2 Purpose
- Maintain and improve quality and yield
BD-CUBE detects process errors early and warns the operator.

- Stabilization of facilities and equipment
Strengthen condition-based maintenance to optimize planned maintenance and reduce breakdown maintenance.

2.1 Utilizing BD-CUBE

1. Detect abnormalities in the process and find the cause. (Online/Offline)
For example, Find undetectable errors by monitoring and control during normal operation

2. Detect abnormal data. (offline)
For example, find abnormal data in one month’s operation data.

3. Detect abnormalities in plant-related equipment in an early stage. (online)
In addition to DCS, find signs of equipment failure in an early stage by additionally registering equipment status (acceleration, strain, current value, etc.).
2.2 Data processing of "BD-CUBE" sign detection system
“BD-CUBE” detects abnormality with high accuracy. It supports efficient analyses of abnormal factors through the impact sensor list displayed on the detailed screen.

2.3 Structure of BD-CUBE sign detection system
Using the process data collected from DCS, BD-CUBE analyzes the sign of “quality and equipment abnormality”.

2.4 Analysis technology of "BD-CUBE" sign detection system
- Features : High accuracy detection technology [Local subspace classifier]
※ The picture of Decision Model is a concept diagram. (Refer to separate technical data for exact information)
MT method (Mahalanobis Taguchi method) | VQC method (Vector Quantization Clustering) | LSC method (Local Sub-space Classifier) | |
Image | ![]() | ![]() ![]() ![]() | ![]() ![]() ![]() |
Technique | parametric technique (represented by the formula) ※Assume normal distribution ![]() ![]() ![]() |
|
|
At learning | (×) Need knowledge of normal distribution. ⇒ It requires 3 ~ 24 times more effort compared to LSC. (It requires sensor extraction and status setting by elapsed time) ※ Calculated by actual Hitachi evaluation | (○) Data can be learned without preparation or effort | (○) Data can be learned without preparation or effort ⇒ 1/3 to 1/24 Man Hour is required compared to that of MT method. ※Calculated by actual Hitachi evaluation |
At evaluating diagnosis | (×) The degree of detection is low and the number of false alarms is not small. (○) It is easy to conceptually understand the meaning of distance. (○) Time required for evaluation is short. | (△) The degree of detection is high and there is little false alarm. (○) It is easy to conceptually understand the meaning of distance. (○) Time required for evaluation is short. | (○) The degree of detection is high and the number of false alarms is small. (△) It is easy to conceptually understand the meaning of distance. (×) Time required for evaluation is long. → Improved by (△) Fast-LSC (Hitachi Technology) |

