Source: Siemens 2023 | Author: Daniel Henningsen
SITRANS SCM IQ enables intelligent monitoring of your machines and mechanical components, such as pumps, gearboxes, or compressors, and detects anomalies in the monitored parts. Utilizing artificial intelligence (AI), SCM IQ learns the normal state of your machine and reliably identifies deviations. Common questions include:
- What type of AI is used?
- How does it work?
- What are the benefits of AI compared to simple thresholds and tolerance bands?
These questions are addressed by explaining the functioning of neural networks (Autoencoder) and their various forms (CNN, LSTM).
Autoencoder: An Autoencoder is a type of artificial neural network used for data compression and reconstruction. It consists of an encoder and a decoder, where the encoder reduces the input to essential features, and the decoder reconstructs the original data from these features, providing the output.
Figure 1: Structure of an Autoencoder (Source: Wikimedia Commons)
The autoencoder learns to compress and reconstruct data autonomously from provided datasets. These datasets do not need additional classifications, a method known as "unsupervised learning," which contrasts with "supervised learning" that requires data classification and input-output pairing. This unsupervised approach simplifies the training phase significantly. The difference between the input and output is described by the reconstruction error.
The autoencoder is trained to analyze the machine's normal state with minimal reconstruction error. A large number of training datasets is necessary to capture all variations of the normal state.
In SCM IQ, the reconstruction error is represented by the Deviation Index. The higher the Deviation Index, the more likely the monitored machine is outside the trained state, indicating anomalous data inputs. The autoencoder in SCM IQ works with vibration data from SITRANS MS200 or general machine data, utilizing different artificial neural networks for analysis. Vibration data is analyzed with a Convolutional Neural Network (CNN) and machine data with a Long Short Term Memory (LSTM).
Convolutional Neural Networks (CNN): A CNN efficiently captures local spatial patterns and is often used for image analysis. It is relevant when the spatial arrangement of data carries information, as with image pixels. Vibration data in SCM IQ comes in packets of 512 time-sequenced, 3-axis measurements, where the arrangement along the time axis and the spatial relationship between measurements are crucial. Unlike traditional threshold monitoring, SCM IQ can assess the relationship between all three axes.
Long Short Term Memory (LSTM): LSTM is a type of Recurrent Neural Network (RNN) with a memory function, influencing current outputs with previous inputs and intermediate results, making it ideal for time-series data analysis. LSTM has a complex memory structure, incorporating short-term and long-term memory. Machine data typically consists of a small number of measurements per dataset, arranged spatially arbitrarily but collected at shorter time intervals, making the relationship between consecutive datasets significant.
In summary, SITRANS SCM IQ provides intelligent monitoring for various machines and components. Its anomaly detection relies on an autoencoder based on either CNN or LSTM.
