Predicting Filter Medium Performances in Chamber Filter Presses with Digital Twins Using Neural Network Technologies

TL;DR

This work presents a neural-network driven digital twin for a chamber filter press to predict key process variables such as pressure and flow rate and to estimate the filter medium lifespan. It compares FFNN and RNN architectures, finding the RNN superior for pressure prediction and the FFNN competitive for flow rate, demonstrating strong generalization across partially known data and reasonable extrapolation to unseen configurations. The digital twin enables real-time data exchange, continuous model updating, and proactive maintenance planning, with potential AR visualization as a future extension. Quantitatively, the approach achieves about $5\%$ relative error for pressure and $9.3\%$ for flow rate on training/validation data, while unknown data show higher errors but retain useful predictive capabilities for operational decision-making.

Abstract

Efficient solid-liquid separation is crucial in industries like mining, but traditional chamber filter presses depend heavily on manual monitoring, leading to inefficiencies, downtime, and resource wastage. This paper introduces a machine learning-powered digital twin framework to improve operational flexibility and predictive control. A key challenge addressed is the degradation of the filter medium due to repeated cycles and clogging, which reduces filtration efficiency. To solve this, a neural network-based predictive model was developed to forecast operational parameters, such as pressure and flow rates, under various conditions. This predictive capability allows for optimized filtration cycles, reduced downtime, and improved process efficiency. Additionally, the model predicts the filter mediums lifespan, aiding in maintenance planning and resource sustainability. The digital twin framework enables seamless data exchange between filter press sensors and the predictive model, ensuring continuous updates to the training data and enhancing accuracy over time. Two neural network architectures, feedforward and recurrent, were evaluated. The recurrent neural network outperformed the feedforward model, demonstrating superior generalization. It achieved a relative $L^2$-norm error of $5\%$ for pressure and $9.3\%$ for flow rate prediction on partially known data. For completely unknown data, the relative errors were $18.4\%$ and $15.4\%$, respectively. Qualitative analysis showed strong alignment between predicted and measured data, with deviations within a confidence band of $8.2\%$ for pressure and $4.8\%$ for flow rate predictions. This work contributes an accurate predictive model, a new approach to predicting filter medium cycle impacts, and a real-time interface for model updates, ensuring adaptability to changing operational conditions.

Figures (11)

• Figure 1: The experimental setup of the chamber filter press, where (a) shows the filter press in its entirety and (b) shows a filter chamber with the filter cloth.
• Figure 2: Architecture of the DT framework for a chamber filter press, illustrating the communication flow between the filter press, real-time measuring techniques, the central database, the predictive model, and the operator.
• Figure 3: Integration of AR with the DT framework TEUTSCHER22. A three-dimensional model of the chamber filter press is overlaid on the real geometry, demonstrating the potential for visualizing operational states and diagnosing issues such as filter cloth wear or system errors in real time.
• Figure 4: Schematic overview of the data acquisition and communication architecture for the chamber filter press system. The setup includes a Delphin data logger connected to sensors for pressure and flow rate measurement, which communicates with a control unit via OPC UA. The control system, represented by a PiXtend V2 board, interfaces with peripheral devices such as a touchscreen and user input terminals through HDMI/USB. Data transfer and remote monitoring are facilitated through a router, enabling Wi-Fi connectivity and LTE-based transmission to a fileserver for storage and further analysis. Developers and operators can access the system remotely or locally for control and data export.
• Figure 5: Comparison of RNN (a) and FFNN (b). Gray nodes represent input neurons, and green nodes represent hidden neurons in both architectures. In (a), the RNN includes feedback loops, represented by the arrows looping back from the hidden neurons to themselves, enabling the processing of sequential data and temporal dependencies. The arrows pointing to the right indicate information flow to the output. In (b), the FFNN consists of direct connections between neurons, without feedback loops, illustrating a simpler network structure designed for static data processing.