Industrial analytics is not magic!

Industrial analytics records and analyzes the data and sensor streams of machines and systems in real time. Using intelligent technologies, time series data can be used to calculate exactly when, for example, the wear on a machine is so high that certain components need to be replaced.

The intelligent networking of machines along the entire production chain is producing ever larger volumes of data in the manufacturing industry, known as "industrial big data". Industrial big data has emerged in the course of the Internet of Things or Industry 4.0 and, in contrast to "conventional" big data, primarily comprises data produced by machines and systems. This data is not only vast in volume, but also completely heterogeneous: temperature measurements, power consumption and pressure in industrial production, performance data from motors, log files from production components to other machines in the process chain or environmental data from machines such as humidity and room temperature.

For companies, this data holds immense potential for optimizing their own production processes and maintenance. In order to leverage this potential, the machine data must be made analyzable and evaluable. And this is where industrial analytics comes into play!

Industrial analytics and predictive maintenance go hand in hand

Industrial analytics records and analyzes the data and sensor streams from machines and systems in real time. Using intelligent technologies, time series data can be used to calculate exactly when, for example, the wear on a machine is so high that certain components need to be replaced. This allows predictions to be made about machine and system component failures or problems and predictive maintenance to be carried out.

Predictive maintenance means nothing other than the predictive maintenance of machines, motors, systems, devices or components. With the help of predictive maintenance, manufacturers, operators or technicians can initiate maintenance measures at an early stage or order spare parts in good time in order to avoid machine or system downtimes or failures. This drastically reduces maintenance costs and improves service in the long term.

In addition to predictive maintenance, industrial analytics can also be used to assess the effectiveness of production processes or optimize production processes by identifying sources of error.

Industrial analytics and systematic data analysis provide operators and manufacturers with a valid decision-making basis for optimizing production processes, enabling them to increase their productivity and gain market advantages.

In order for industrial analytics to be used successfully, the right question is crucial: What do you want to find out? What is the cause of a problem? How urgently does it need to be rectified? What spare parts are needed?

The right questions are much more important than algorithms.

In order to be able to ask the right questions, the machines, devices and components to be examined and their processes must be understood. Only then can the appropriate analysis method and the corresponding algorithms be selected.

Let's take the example of a broken exhaust valve in a diesel engine. This results in serious consequential damage to the engine, including destruction of the cylinder head and ultimately the turbocharger.

This damage can be predicted and prevented with the help of industrial analytics. This requires a basic understanding of the role the valve plays in the combustion process: If the valve is damaged but not destroyed, it can no longer seal the combustion chamber perfectly. As a result, the combustion process is less efficient.

This reduced efficiency is reflected in a drop in exhaust gas temperature. The good news with modern engines is that the exhaust gas temperature is measured by sensors, i.e. this data is always available and can be used to identify problems.

This makes it possible to determine the premise for the data analysis: The detection of a deviation in the exhaust gas temperature. The clear formulation of the problem ultimately makes it easy to select the appropriate algorithm.

This requires a mathematical function to determine the expected normal exhaust gas temperature. There are also simulation models for such calculations, but these require the experimental quantitative determination of physical process constants. In practice, this is far too expensive and time-consuming. Therefore, the data of the actual process in the engine is used.

In the simulation, the constants represent the individual properties of a specific motor. This information is implicitly contained in the actual data of the variable parameters of a motor. The adjustment of the simulation model results in a simplified empirical predictive model for estimating the expected exhaust gas temperature of the engine.

The input parameters include the charge air temperature, the engine speed, the output energy produced and the ambient temperature. The latter must withstand the temperature changes at night and during the day. All this data is provided by the engine monitoring system.

Further metadata is required for a complete analysis:

Device metadata - type, year of manufacture, etc.
Sensor and message metadata - it is important to note here: A temperature sensor on an American machine may give its measurements in Fahrenheit, while a German version does so in Celsius. The same error is often reported using different error codes, depending on the year of manufacture and/or different firmware versions. Thus, sensor and message metadata enable the normalization and standardization of measurement data.
Business metadata - depending on the maintenance business model, different maintenance contracts exist for different machine operators or owners, e.g. different availability of remote monitoring and reporting services. This information must also be taken into account for certain devices.

To summarize, the prerequisites for the successful use of industrial analytics are

• Understanding cause and effect
• Identification of a suitable predictive model (predicting normal values enables the detection of abnormal values)
• Availability of the required input data
• Selection and application of the appropriate algorithm