SPHERA (Systems Prediction and Health Rate Analysis) is the artificial intelligence solution proposed by Project Consulting for predictive maintenance and optimization of data center resources. SPHERA uses current monitoring systems to collect data from agents and sensors located in the data center.
Using a mixed technology based on analytics, machine learning and deep learning algorithms, SPHERA analyzes all the data acquired by the monitoring systems, extracting the useful information for optimizing the systems maintenance processes and energy consumption.
The operation pipeline is indicated in the following diagram:
The data relating to personal data, operation, consumption and events are acquired by the monitoring systems. The first phase of analysis carries out the preprocesses all the data, in order to verify its integrity, semantic correctness and robustness. The next phase is able to identify any correlations between events, system parameters and consumption. This phase involves the use of supervised and unsupervised AI algorithms, which use all the available information, both historical and current. The objective of this phase is to classify and cluster data and prepare data sets for training machine learning algorithms, which will be used in the next phase
Once the representative datasets of the data center operation have been identified, they are given as input to a machine learning algorithm for the self-learning of anomalies and malfunctions, and for the optimization of operational and / or production processes. Once trained, the system will be able to propose autonomously and automatically the operating parameters values, according to an order of priority related to the criticality, the urgency of some operations, or the correlation with other parameters or other factors of weight, issued from the “intelligent analysis”.
This operation mode makes the use of existing monitoring systems particularly effective.
SPHERA exclusive features:
- Dynamic measurement of the events and parameters correlation within the same device/system and between different devices / systems
- Prediction of individual parameters, based not only on their historical performance but according to an analysis that takes into account the correlations, identified independently by the machine learning engine
- Availability of a priority actions list based on the data center effective operation (automatically proposed by the machine learning engine)
- Measurement of the systems and subsystems functioning quality and their future estimate on the basis of the dynamic clustering produced by AI algorithms
- Measurement of the change in the quality of current maintenance based on the actions performed based on predictions
- Measurement of current energy consumption and its prediction
Despite the complexity of managing big-data from monitoring systems and their analysis with AI algorithms, the solution produces simple and intuitive information by displaying dynamic priorities independently identified by the machine learning engine. Once acquired this knowledge, it will be possible to navigate between devices and systems using the current monitoring systems. SPHERA does not have complex GUIs and therefore does not require additional training costs. In order to operate SPHERA only needs to have access to the data channel of the current monitoring system.
SPHERA system can be used for transfer learning on synthetic data specially created for predictive simulation. This is a particularly effective point when you want to simulate particular operating conditions within the data center that have never occurred before, for example to predict the consequences of a failure or a peak workload. In this sense, SPHERA can become an important tool for planning future expansions of the data center or its replicas. The introduction of Sphera determines an enhancement of the current monitoring systems, with consequent enhancement of the asset.scheda-sphera