Abstract:Soft sensors are crucial in bridging autonomous systems' physical and digital realms, enhancing sensor fusion and perception. Instead of deploying soft sensors on the Cloud, this study shift towards employing on-device soft sensors, promising heightened efficiency and bolstering data security. Our approach substantially improves energy efficiency by deploying Artificial Intelligence (AI) directly on devices within a wireless sensor network. Furthermore, the synergistic integration of the Microcontroller Unit and Field-Programmable Gate Array (FPGA) leverages the rapid AI inference capabilities of the latter. Empirical evidence from our real-world use case demonstrates that FPGA-based soft sensors achieve inference times ranging remarkably from 1.04 to 12.04 microseconds. These compelling results highlight the considerable potential of our innovative approach for executing real-time inference tasks efficiently, thereby presenting a feasible alternative that effectively addresses the latency challenges intrinsic to Cloud-based deployments.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the challenges such as network latency, data security, and low transmission rate faced by traditional cloud - based soft - sensor deployments in real - time fluid flow estimation. Specifically: 1. **Network Latency and Reliability**: Cloud - based soft - sensors rely on a stable network connection. Any network instability will significantly delay the transmission and processing of sensor data, affecting the performance of real - time applications. 2. **Data Transmission Rate**: The low transmission rates (0.3 kbps to 50 kbps) of many wireless sensor networks (such as LoRaWAN) make it difficult to continuously and real - time transmit a large amount of sensor data to the cloud. 3. **Data Security**: Transmitting data to the cloud may bring data privacy and security issues, especially in applications involving sensitive information. To solve these problems, the paper proposes a new method - **device - side soft - sensors**. By directly deploying artificial intelligence (AI) on devices in wireless sensor networks instead of relying on cloud services. This method not only improves energy efficiency but also enhances data security and the real - time performance of the system. In terms of specific implementation, the paper combines the microcontroller unit (MCU) and the field - programmable gate array (FPGA), and utilizes the fast AI inference ability of FPGA to achieve real - time fluid flow estimation. The experimental results show that the inference time of FPGA - based soft - sensors ranges from 1.04 µs to 12.04 µs, which is far superior to the MCU platform and has lower energy consumption, demonstrating the great potential of this method in practical applications. ### Formula Summary - **Mean Squared Error (MSE)**: An indicator used to evaluate model performance. \[ \text{MSE}=\frac{1}{n}\sum_{i = 1}^{n}(y_i-\hat{y}_i)^2 \] where \(y_i\) is the true value, \(\hat{y}_i\) is the predicted value, and \(n\) is the number of samples. - **Learning Rate Scheduling**: Adjust the learning rate during the training process to optimize model performance. \[ \text{learning\_rate}=\text{initial\_learning\_rate}\times\gamma^{\left\lfloor\frac{\text{epoch}}{\text{step\_size}}\right\rfloor} \] where \(\gamma\) is the decay factor, \(\text{step\_size}\) is the step length, and \(\text{epoch}\) is the current round number. Through these improvements, the paper demonstrates the efficiency and feasibility of device - side soft - sensors in real - time fluid flow estimation, providing an innovative solution to solve the existing problems of cloud - based deployments.

On-Device Soft Sensors: Real-Time Fluid Flow Estimation from Level Sensor Data

Towards Auto-Building of Embedded FPGA-based Soft Sensors for Wastewater Flow Estimation

FlowPrecision: Advancing FPGA-Based Real-Time Fluid Flow Estimation with Linear Quantization

Soft-Sensor Development Through Deep Learning with Spatial and Temporal Feature Extraction of Complex Processes

Energy-efficient Pipelined DTW Architecture on Hybrid Embedded Platforms

Hardware–Software Partitioning for Real-Time Object Detection Using Dynamic Parameter Optimization

Adaptive Stream Processing on Edge Devices through Active Inference

A Survey on Deep Learning for Data-Driven Soft Sensors

Exploiting Real-Time FPGA Based Adaptive Systems Technology for Real-Time Sensor Fusion in Next Generation Automotive Safety Systems

A Hardware Realization Framework for Fuzzy Inference System Optimization

A Device-on-Chip Solution for Real-Time Diffuse Correlation Spectroscopy Using FPGA

TFT-Based Near-Sensor In-Memory Computing: Circuits and Architecture Perspectives of Large-Area Edram and ROM CiM Chips

DIFT: Dynamic Iterative Field Transforms for Memory Efficient Optical Flow

A Future Perspective on In-Sensor Computing

A Bi-Directional Co-Design Approach to Enable Deep Learning on IoT Devices

Context-Aware Collaborative-Intelligence with Spatio-Temporal In-Sensor-Analytics in a Large-Area IoT Testbed

In-Sensor Computing Realization Using Fully CMOS-Compatible TiN/HfO x -Based Neuristor Array

Development of soft sensor by incorporating the delayed infrequent and irregular measurements

A Novel Single Chip Implementation of MEMS INS Data Acquisition and Processing System Using FPGA and Its Soft Processors

SciFlow: Empowering Lightweight Optical Flow Models with Self-Cleaning Iterations

Real-time FPGA Implementation of an Optic Flow Algorithm Suitable for Background Oriented Schlieren Technique