Published Date 08/02/18 07:30
The increasing tendency to have connected devices that provide remote access to installations through monitoring systems is paving the way for the systematic collection of vast amounts of data. In the HVAC/R sector in particular, unit variables, control and operating parameters, such as valve position, set point, compressor status, water/air temperatures and so on, if appropriately recorded can represent a very useful source of information to determine unit performance and operating conditions.
This important result has been achieved thanks to the spread of IOT systems. In HVAC/R applications, these networks comprise sensors or controllers that are natively connected, i.e. capable of transferring the main information via open-standard protocols. Local edge computing systems are often used to centralise data, execute on-field logic and guarantee connectivity to enterprise systems. These edge systems may be simple gateways or even smart supervision systems. The aggregated data, finally, are brought together on enterprise systems that, installed either on the premises in local servers or available on the cloud, can perform complex processing operations thanks to their virtually unlimited computational capabilities.
The new challenge then is to exploit this huge amount of data to create new value. To date, various data processing techniques are available, with the ultimate goal of streamlining processes and enabling new business by simplifying what without the appropriate information would otherwise be very complex.
The most common terms used in this new field of Intelligence of Things are:
|Artificial intelligence || |
| ||Problem-solving techniques through simulation of the cognitive functions of the human brain, including logic, perception and prior experience; |
| || |
|Machine learning || |
| ||The ability of computers to identify, through the application of algorithms, recurring patterns in historical sequences of collected data; |
| || |
|Predictive models || |
| ||Thanks to the relationships between the input data and the target data identified through the analysis of historical data, these models are able to predict the expected behaviour of target variables based solely on new input data; |
| || |
|Deep learning || |
| ||Use of neural networks, often with a high number of layers, to create very complex predictive models. T his today is the most advanced AI technique.|
In HVAC/R applications, the main focus of the application of these technologies today is site maintenance and optimisation. In complex equipment such as that used in this sector, where system operation is profoundly affected by surrounding conditions (outdoor/indoor humidity, seasonal configuration...), in order to provide a deep understanding of the actual performance and operating behaviour of a mechanical/energy system, a non-linear model is needed that correlates a vast set of variables.
Such analysis can offer highly significant information, allowing the various decision makers involved to make informed decisions with more certain outcomes. In particular, some significant uses may be:
|Predictive maintenance || |
| ||Thanks to dynamic alerts generated in the event of unit behaviour (e.g. energy consumption) that deviates from the predictive model, preventive maintenance can be performed to avoid malfunctions in the field. Such services ensure continuous operation of mission-critical systems, which is essential in industrial applications. In addition, maintenance costs are reduced, as actions that were previously taken in response to an emergency can now be planned in advance, thus transforming urgent service call-outs into routine maintenance. |
| || |
|Performance optimisation and what-if analysis |
| ||The model identifies the configuration that maximises efficiency, precisely quantifying the benefits in terms of energy savings. Calculation of the total energy saved through the difference between the consumption actually measured and the consumption that would have been expected had the optimisations not been applied, makes it possible to highlight the benefits obtained. This is often fundamental in order to sell a remote control or supervision system. |
| || |
|Benchmarks between different units || |
| ||In order to identify the best configuration or the best maintenance process, the unit’s performance is compared against what is predicted by the model. Such information is very complex to obtain, and is based on analysis of the data from individual units. Environmental conditions in fact have a significant influence on unit performance and energy consumption. Only by creating a predictive model can external variables be “normalised”, making it possible to determine whether the unit is behaving better or worse than others operating in the same conditions |
These are just some examples of practical applications of the enormous possibilities available by processing the collected data. The more these new techniques are refined and applied to real cases, the more the Intelligence of Things will become the driver of new business opportunities.
Machine Learning Applied to HVAC Systems for Energy and Comfort Management; Energy Way Srl
Controlling machine-learning algorithms and their biases; McKinsey&Company
IoT @Bottle Cooler
IoT: data processing in a world of connected machines