Science logo

Science and engineering applications

By Roberto Lopez, Artelnics.

Predictive analytics in science and engineering is undergoing an impressive growth. More and more companies are adapting that technique into their daily operations, not only to better manage the present, but also to increase the probability of future success.

Some opportunities here are the following:

Output prediction picture

1. Output prediction

Collecting input and output values from a plant can be used to forecast its response, reducing operational costs, and optimizing management resources. Predictive analytics allows to include all types of internal and external variables, and provides higher accuracy in this type of applications than traditional techniques.

An example is the accurate prediction of power generated at a combined cycle plant. Neural networks might be used here to learn the solar power as a function of weather conditions and panel characteristics. Accurate forecasts here will allow an efficient management of resources within the electric company. The most relevant variables for the creation of solar power can be classified in the following groups:

  • Input variables:
    • Temperature
    • Exhaust vacuum
    • Ambient pressure
    • Relative humidity
  • Target variable:
    • Net hourly electrical energy output

Output prediction dataset picture

The neural network defines the model mentioned before. It takes the ambient variables and produces the corresponding electrical energy output. The following picture shows a graph of the deep learning architecture for this example.

Neural network graph

From a data set containing information about the above variables, Neural Designer will create the corresponding predictive model.

Performance optimization picture

2. Performance optimization

Tracking the behaviour of a system enables to operate it at peak level. Improvement can have different flavours: reduced consumption, increased velocity, diminished noise, etc.

An example is to optimize the behaviour of an engine, based on the most important operational variables. Different objectives can be considered here, which include power improvement and emissions reduction. The data set used in this study contains quantitative information about the fuel rate and angular speed of an engine, and its corresponding torque and N2O emissions. The data set for this application is illustrated below.

  • Input variables:
    • Fuel rate
    • Angular speed
  • Target variables:
    • Torque
    • Nitrous oxide emission
Performance optimization dataset picture

Once the model has been designed, we can use it to predict the torque and the emissions as a function of the rotational speed and the fuel flow. For that purpose, we can use the mathematical expression of the neural network, which is listed next.

				scaled_fuel_rate = 2*(fuel_rate-0.6)/(314-0.6)-1;
				scaled_speed = 2*(speed-612.1)/(1801.8-612.1)-1;
				y_1_1 = tanh(-1.11096-1.61239*scaled_fuel_rate+0.430536*scaled_speed);
				y_1_2 = tanh(0.53813+0.546026*scaled_fuel_rate-3.48146*scaled_speed);
				y_1_3 = tanh(2.41882-2.12554*scaled_fuel_rate+3.45315*scaled_speed);
				y_1_4 = tanh1.31628-1.76257*scaled_fuel_rate-0.881265*scaled_speed);
				y_1_5 = tanh(-0.134698-0.138303*scaled_fuel_rate+2.60122*scaled_speed);
				y_1_6 = tanh(0.110113+0.472119*scaled_fuel_rate-0.0221803*scaled_speed);
				scaled_torque = (-0.190618-0.174056*y_1_1+0.215817*y_1_2-0.00182521*y_1_3-0.0968754*y_1_4+0.287748*y_1_5+1.61218*y_1_6);
				scaled_nitrous_oxide_emission= (-0.780686-1.56753*y_1_1+1.44214*y_1_2-0.406431*y_1_3+0.453272*y_1_4+1.93066*y_1_5-1.23584*y_1_6);
				nitrous_oxide_emission = 0.5*(scaled_nitrous_oxide_emission+1.0)*(1774-0)+0;

Predictive maintenance picture

3. Predictive maintenance

Predictive maintenance looks for anomalies, i.e., measurements that are unexpected and therefore might indicate a problem - but are not yet so severe that they are a failure. This approach makes companies save money, since they will have shorter downtime and less lost production, better planning of people and materials and reduced repair costs. Neural Designer has been conceived to build complex predictive models that represent a broad range of variables associated with the failure of equipment.

An example is to spot potential failures of wind turbines earlier, therefore allowing to fix them before they happen. To do that, neural networks analyze sensor data from outside and inside the turbine: atmosphere, blades, rotor, brake, shaft, gearbox, generator, yaw, tower, etc. If during operation they detect an unusual pattern, a flag is raised. The process begins by creating a data set with variables that affect the operation of a turbine:

				-External variables: Wind speed, wind direction, air temperature, etc.
				-Internal variables: Blade pitch position, rotor angular position, brake pad position, 
									 shaft deflection, gearbox lubrication level, generator rpm, 
									 yaw position, output power, etc. 

The so called auto-association model learns the nominal operation of the wind turbine at given conditions. If the current operation does not associate with its expected operation, there is an anomaly. In this case, an analyst can investigate the cause and decide on a course of action together with the operations staff at the wind farm.

 Quality improvement picture

4. Quality improvement

A main solution within predictive analytics is the continuous improvement of quality, which is designed to obtain products with the best specifications. This is an iterative process, in which the company collects manufacturing data, creates a predictive model of the product quality and uses it to design the next manufacturing process.

An example is the design of concretes having specific properties by selecting their constituent materials and optimizing the proportions. The result is a product with the highest quality by following the specifications and reduced cost by using the exact mix.

Quality improvement dataset picture

The neural network defines the model mentioned before. It takes the proportions of the constituent materials and produces the corresponding concrete properties. The next plot shows the output compressive_strength as a function of the input water. The x and y axes are defined by the range of the variables water and compressive_strength, respectively.

Network architecture for quality improvement of concrete
Directional line chart.

Fault detection picture

5. Fault detection

Automatic fault detection concerns with monitoring a system, identifying when a fault has occurred, and pinpointing the type of fault and its location. This is always a big concern in industry production. The higher quality a production is required to have, the better fault diagnosis method the factories should apply. Two approaches can be distinguished here: (i) A direct pattern recognition of sensor readings that indicate a fault and (ii) an analysis of the discrepancy between the sensor readings and expected values.

An example is to recognize patterns of different steel plates faults, based on all kinds of image attributes. The variables used for this application set are the following:

				-Independent variables (attributes): 

				-Dependent variables (faults): 

The main outcome from this project the improved accuracy in the detection of steel plate faults. That means a high reduction of direct and indirect costs for the factory.

Activity recognition picture

6. Activity recognition

Activity recognition aims to recognize the actions and goals of one or more agents from a series of observations on the agents' actions and the environmental conditions. Since the 1980s, this research field has captured the attention of several computer science communities due to its strength in providing personalized support for many different applications and its connection to many different fields of study such as medicine, human-computer interaction, or sociology.

An example is to identify the type of movement of an individual knowing the information of the accelerometer of their mobile phone. The data set will have as dependant variables the velocity, the high and the directional movement of the mobile. The output is the kind of movment of the individual.

The variables will be the next ones:

  • Input variables:
    • Body acceleration
    • Gravity acceleration
    • Body acceleration jerk
    • Body angular speed
    • Body angular acceleration
    • Body acceleration magnitude
    • Gravity acceleration magniture
    • Body acceleration jerk magnitude
    • Body angular speed magnitude
    • Body angular acceleration magnitude
  • Target variables:
    • Walking
    • Walking upstairs
    • Walking downstairs
    • Sitting
    • Standing
    • Laying Down

After analysing the data, the model has produced the next confusion matrix:

Neural network

The diagonal values are the correctly predicted instances number, the rest of the matrix represents the misclassified ones. As we can see, the model has predicted accurately a high number of cases.