Visual and Data Analytics Tool User Guide
The tool is accessible through the EFPF portal only to the users that have access to the corresponding data sources that the tool analyze and visualize. The tool is delivered to the users as type of dashboard that provides three main options: (1) visit the real-time monitoring dashboard, (2) visit the analytics tab and (3) visit the administration tab in order to add new data to database.
The monitoring dashboard provides real-time visualizations for machine’ condition or for sensors’ data. An example for fill level sensors monitoring view is available in next figure:
The analytics tab that is the main view of the Visual Analytics tool provide to the user the following options:
The first option enable the user to choose the type of analysis that he/she wants to perform, the second option is to select a raw material that want to load relevant data from a database in order to analyze them or skip this option and choose the third option that enables to load data from csv file.
As soon as the user defines, the type of analysis he/she wants to perform and deploy the corresponding data then the following options/interfaces are available:
In 1st step the user is able to select the type of graph that he prefers to show the data that just load. In 2nd step the user is able to see the loaded data in a table and in 3rd step is able to get the data representation that it was selected on step number 1. In 4th step the user is able to select the data analytics methodology and on 5th step the diagram type that he wants to get the analytics output. The output is available on 6th step.
The following analytics methodologies are available to the user:
Machine Vibration Diagnosis Profile applied on a Polishing Machine
- Dynamic solution based on real-time data coming from deployed vibration sensors. The method creates a profile of machines normal operation
- Real-time detection of abnormal vibrations by detecting the time point(s) when abnormal vibrations occur from the profile of the eigenvalues’ sums
- The basic assumption of MVDP is that significant eigenvalue sums with simultaneous significant variations could point out to abnormal vibrations
- Maintenance manager visually informed via the Visual Analytics when the machine’s activity surpasses the abnormal vibration threshold
Fill Level Trend Analysis
The fill level sensors data that are available in Circular Economy pilot analyzed in order to provide estimations that can optimized planning procedures. A Trend Analysis methodology is available. The key aspects of the deploying solution that are available to the user through EFPF portal are the following:
- Real-time analysis of fill level sensors data
- Trend Analysis applied in order to create a profile for fill level trend
- Slope Statistic Profile method is applied on the time series of recordings (percentages) of a fill level sensor
- Waste management company is able to define which bin has the most aggressive trend in order to arrange a pick-up
Methodologies for Tonnage Forecasting
The moving average algorithm aims to predict future values for the time series by finding an n-length series of values (where the value of n is chosen by the user from between 2 to 5) that, when repeated for the whole length of the time series, best fit the time series. To compute these n values the time series is split into n-lenght segments and the average value of each of the n points is calculated. The prediction is made by repeating this series of average values for the required number of values. In order to deal with time series that display a trend, the moving average value of the time series is computed and from this the average trend is computed. This trend is then removed from the original time series, the moving average prediction is calculated, and then the trend applied to the prediction.
A n-order auto-regression model provides a prediction for a time series by predicting future values as a linear sum of the n previous values.
We used the function ar_model.AR from the python library statsmodels.tsa to create the auto-regressive model. The function uses the conditional maximum likelihood estimation method to automatically determine the best order for the model and compute the model’s parameters. In order to deal with seasonalities in the time series, the user can apply a de-seasonalization method by choosing the length of a seasonality period (e.g. 12 for yearly seasonality for monthly data). In that case, the algorithm computes the average value for each point in the period, subtracts those values from the time series, computes the auto-regression model, uses it to predict the future values of the time series and finally adds the seasonal averages into the predictions.
The Markov chain prediction algorithm uses the time series values to compute a transition matrix between three different time series states: value increases, value decreases, and value remains constant. Using the transition matrix the algorithm can not only compute the probability of the time series moving from one state to another (e.g. the probability of the value increasing after decreasing in the previous time step), but also the probability of the time series moving from one state to another after any number of time steps. The method does not provide the user with specific prediction for future steps of the time series, but it can provide a prediction for general movement of the time series in an easy to understand format while also providing specific statistical probabilities. Moreover, the Markov chain method allows the user to produce effortlessly the probabilities for the time series’ movement over a large number of time steps, again providing precise and easy to understand statistical probabilities.