Data Science

The 9 Artificial Intelligence trends 2020 to keep an eye on – pt 2

We have already read the first 4 of the 9 trends in the world of artificial intelligence for 2020.

We continue the list with other 5 applications of this technology, hoping you will find some ideas to put artificial intelligence at the service of business goals and plant performances.

5. Interception of energy inefficiencies in consumption or production

This application is already known to the Italian market, pioneer in applying energy models for  quickly understanding if you are consuming well or not (due to the energy cost, among the highest in Europe).

Using Machine Learning algorithms, that are more complex than linear regression (which is still a great start and it learns from data in the same way as any other machine learning algorithm), it is possible to characterize with more precision the optimal energetic behaviour of an asset, a building, a plant. The aim is to be able to quickly identify deviations from normal behaviour and act accordingly.

It is estimated that by applying energy consumption models, it is possible to save between 1% and 10% on energy costs.

(Here you find our case study where applying energy models, we reduced the energy consumption of a production line by 11%).

Similarly, it is possible to control through intelligences the plant producibility, whether it is a small co-generator or a large combined cycle power plant, up to renewable wind and photovoltaic farms.

Digital maturity required: defined –

• – Models should receive data every week at least
• – It is necessary to have a historicized database appropriate for the training of the models
• – A hardware and software architecture is also needed to support the communication flow and the installation of an AI platform.

6. Precise baselines for savings quantification

The necessity to demonstrate the savings achieved for activating EPC contracts or for the possibility to receive incentives, has stimulated the market to provide precise solutions by using linear regression and normalization of consumption.

With more complex algorithms, the baselines can be more precise and include several variables, also related to each other, to evaluate the status of the system. As a result, the comparison is always better representative of the savings actually achieved thanks to the operations of efficiency or extraordinary maintenance, as the factors that influence consumption variability are already included in the assessment.

Digital maturity required: managed –

• – It is necessary to have a precise database historicized and representative of the ex-ante status.
• – It is indispensable to collect data that are useful for the use of the model related to the ex-post status. These data do not need to be directly connected to the model, as the evaluation is done at fixed intervals and can include the manual data upload.

7. Prediction of production capacity and consumption forecast

The energy market requires energy producers to forecast the maximum plant producibility for the next day, while consumers need the forecast of the probable energy demand for the same period. Often, finding a model that represents both is difficult and the result is not precise, thus conservative coefficients are applied to avoid sanctions.

By training machine learning models on the basis of the history of consumption or production in accordance with production volume forecasts, weather conditions and other available information, it is possible to increase the accuracy of the model, guaranteeing higher margins to resellers and less expenses to buyers.

Digital maturity required: managed –

• – It is necessary to have a precise database historicized and representative of the maximum load for the production and of the normal consumption curve for the user, combined with the forecast for the same periods.
• – The use of the solution can be done by manual upload. However, having a solution that can make the prediction through data provided automatically and generate reports ready for sharing, simplifies the activity a lot.

8. Process optimization

As well as quality control through image recognition, it is possible to actively act on processes and plants by leveraging machine learning potentialities.

Creating a model able to represent the system behaviour allows to minimize production costs and maximize efficiency by asking the solution to autonomously act on the control parameters of the process (such as set points, temperatures, pressures, flows etc.).

The difference from a normal controller is how the result of the target function (minimum cost or maximum efficiency) is identified, as it is based on how the asset would have behaved under certain conditions. The model becomes a simplification of the digital twin of the process or of the asset, precisely representing its behavior basing on the different operational conditions.

Digital maturity required – digital oriented

• – The system has to be integrated, connected in real-time with the process data and able to act precisely on set-points and control parameters.
• – The amount of data needed to train a sufficiently precise intelligence is high and it has to be updated periodically in order to always represent the correct plant functioning.

9. AI on edge

It sounds like the title of a science fiction movie, but it is nothing more than an intelligent sensor capable of operating artificial intelligence algorithms directly on site. Whether it is anomaly detection, predictive maintenance, performance control or others, having the intelligence in the field saves data traffic, space in central servers and cybersecurity problems caused by data exposure.

The intelligence runs directly on the hardware installed on site and can interact with operators through interfaces and other signals.

Digital maturity required: integrated and interoperable –

• – The intelligence has to be trained according to historical data
• – It is necessary to have a certain control over the intelligence performance in order to avoid degradation of the system reliability.

 

The spread of Artificial Intelligence in the industrial world is already reality: intelligences can be implemented into industrial processes, maintenance management, systems control and much more. Actually, not all the applications require a highly integrated system. Each one of the applications presented can already  be implemented and many are already examples of success.

If you want to participate to the industry revolution thanks to AI, contact us to discuss about it together! We can also evaluate your company’s digital maturity and advise you on which projects to develop and how.

The 9 Artificial Intelligence trends 2020 to keep an eye on – pt1

Artificial Intelligence, protagonist of the scientific debate in the recent years, is the technology that is revolutionizing our industry, as electricity did more than 100 years ago.

While in the field of services, finance and medicine, AI techniques have already been applied for years, in the industrial sector they have recently come to support operators in O&M activities, process control and energy efficiency.

The growing presence of a solid database, which is the result of the IoT investments made in the past years, and of new technical standards defining how and what to monitor – combined with increased market awareness – open the way to new solutions and applications.

As in all fields, it is very easy to make long-term predictions about the spread of artificial intelligence. How close is the future world where machines will be autonomous and thanks to artificial intelligence the human supervision will not be necessary any longer?

Rather, what is already in the final stage of experimentation or is already available and can be applied by Italian and international companies with a sufficient digital maturity? (To evaluate the digital maturity of your company, you can fill in the simple assessment questionnaire that we have created covering different business areas: request it here).

Below, we look together at 9 Artificial Intelligence applications that are already reality in 2020.

1. Predictive and on-condition maintenance for processes, assets and plants

Predictive maintenance through artificial intelligence is the main focus of AI applications in manufacturing. Many international big players are investing on AI and predictive maintenance in order to digitalize processes and support the O&M teams in the management of failures and anomalies. Thanks to the increasing amount of data and the efforts to standardize databases, these technologies become more and more accessible.

The fundamental limit encountered so far is the comprehensibility of the results and the models industrialization. These two problems can be solved through the use of intuitive user interfaces (which require to properly study the process and interpret the results), and the installation of the solutions in cloud. This way, AI-based predictive maintenance solutions will be increasingly accessible to teams that operatively need information, without the need to understand control charts.

(We have already written some case studies on this topic: you can find them here)

Digital maturity required: integrated and interoperable –

• – Models must receive data at intervals of not more than 1 hour
• – It is necessary to have a historicized database adequate for the models training
• – A hardware and software architecture is also needed to support the communication flow and installation of the AI platform (you can look at our solution for industrial IoT here).

2. Image recognition for quality control

Whether it is manufacturing of precision mechanical parts, seals, or bottle caps, all process lines have to ensure a low percentage of scraps. Above all, they need to be integrated in order to automate the recognition of parts under defined quality standard.

Often, screening techniques alone are not enough to find a small scratch or a wrong element. This is why some companies are evolving their quality control algorithms to include intelligent models that can learn by  themselves which pieces are good and which have to be discarded.

Thanks to machine learning, quality control mechanisms increase their accuracy, ensuring product quality with less margin of error.

Digital maturity required: oriented to the digitalization – 

• – It is necessary to have a good number of historicized pictures of good and defective parts. With the correct information the intelligences can learn to recognize the features that make a piece flawed
• – It is necessary to have or install a controller that can support machine learning on edge and coordinate the procedure to report the defect consequently.

3. Anomalies detection through image recognition

Predictive maintenance is often not easily applicable in many and various causes: from the difficulty of installing sensors in some areas to the data property issues (as they have those companies that provide only maintenance services), up to the difficult representation of the physical phenomena behind the fault.

For some cases, however, it is possible to apply images recognition algorithms to identify assets and plants problems. It can be through a photo or an image from a thermal camera.

A maintenance specialist can understand if there is a problem just by looking at certain types of assets (filters, pipes, pictures of panels, etc.). By training intelligences on the basis of expert technicians know-how, it is possible to implement a model capable of doing the same.

With the same technology with which Facebook recognizes the faces of your contacts or advanced clinical diagnosis are made, it is possible to identify visible anomalies in the assets and automatically send a request for a detailed audit, also by new maintainers who do not have the same experience.

Digital maturity required: managed –

• – It is necessary to have a trained technician to engage in the initial part of training and a database of asset photographs
• – With a good system of asset management and CMMS it is possible to structure a database sufficient for the purpose

The solution can be connected directly to the CMMS, and its use does not require a continuous dataflow, hence the implementation of AI for the identification of visible anomalies is relatively simple.

4. Anomaly detection and data quality

The base of artificial intelligence is the data, quality data, that has to be controlled. It is however typical of field data to be incomplete, blocked or abnormal. Thanks to artificial intelligence, it is now possible to spot and process strange values.

Standard cleaning procedures risk not to detect some anomalies while intercepting others that are not real anomalies. Artificial Intelligence on the other hand can understand what is the correct trend of data from the meter under control. Solutions of this type, installed on edge or in cloud, guarantee a data truthfulness that exclude problems of peaks, freeze or misalignment of the measurement compared to the normal values.

 Digital maturity required: initial –

• The intelligence is autonomously able to learn and it could be virtually installed also on the home counter. Clearly, an organization that requires a technology like this has reached a fairly high digital maturity, which leads it to want to ensure data quality quickly and reliably.

 

These are the first 4 trends of artificial intelligence in 2020, on Monday we will publish the following 5 to understand how to put AI at the center of industrial processes.

Digital maturity levels:

• Level 1: Initial
• Level 2: Managed
• Level 3: Defined
• Level 4: Integrated and interoperable
• Level 5: Digital Oriented