As an indispensable component of modern energy infrastructure, solar panels have experienced rapid deployment globally. However, the frequent occurrence of extreme cold weather events poses significant operational challenges, particularly snow accumulation on panel surfaces. This issue severely compromises the power generation efficiency of solar panels, leading to substantial economic losses and even catastrophic structural failures due to excessive snow loads. To address this critical problem, this thesis presents a comprehensive investigation into the design, optimization, and implementation of a snow coverage monitoring and control system for solar panels, utilizing the principle of interdigitated capacitive sensing technology.
The increasing reliance on solar panels for clean energy generation necessitates robust monitoring solutions to ensure operational reliability in harsh winter conditions. Traditional monitoring methods have demonstrated limitations in accurately detecting the presence and thickness of snow on solar panels. This research focuses on overcoming these limitations by developing a high-precision, cost-effective monitoring system. The working principles of traditional parallel-plate capacitors and advanced interdigitated capacitive sensors were meticulously analyzed. Leveraging the capabilities of COMSOL finite element analysis software, a detailed three-dimensional geometric model of the interdigitated capacitive sensor was constructed. This model served as a virtual testbed for comprehensive single-factor simulation analyses and orthogonally designed experiments, aimed at dissecting the influence of key structural parameters on sensor performance. The experimental findings clearly indicated that for output capacitance and sensitivity, the factors ranked in descending order of influence were: interdigital finger length, number of interdigital finger pairs, finger spacing, and the electrode metallization factor. Conversely, for the maximum detection distance, the primary influencing factors were ranked as: electrode metallization factor, finger spacing, number of finger pairs, and finger length.
A significant contribution of this work is the development of a novel multi-objective optimization method, termed IGOOSE-LSSVM-NSWOA, designed to globally optimize the trade-offs between sensor size and output capacitance. This advanced approach enhances the foundational GOOSE algorithm by incorporating several sophisticated strategies: Piecewise chaotic mapping for superior initial population diversity, a golden sine strategy to improve global exploration, and a nonlinear dynamic adaptive inertia weight factor to balance exploration and exploitation effectively. These enhancements considerably strengthen the algorithm’s ability to escape local optima and improve its overall optimization performance. The improved IGOOSE algorithm was then intelligently employed to automatically optimize the internal parameters of a Least Squares Support Vector Machine (LSSVM) model, thereby eliminating the need for manual, often suboptimal, parameter tuning. The resulting IGOOSE-LSSVM predictive model serves as an efficient and highly accurate surrogate for the computationally expensive finite element simulations. This model was subsequently integrated as the objective function within a Non-dominated Sorting Whale Optimization Algorithm (NSWOA) to perform multi-objective optimization. The experimental results demonstrated the superiority of this joint optimization approach; the IGOOSE algorithm outperformed the standard GOOSE algorithm in both robustness and global search capability, while the IGOOSE-LSSVM model achieved a coefficient of determination (R²) of 0.9951, significantly surpassing other models. Through this rigorous optimization process, the optimal combination of structural parameters for the interdigitated capacitive sensor was identified: a metallization ratio of 0.48, an interdigital spacing of 0.33 mm, a finger length of 47.46 mm, and 9 pairs of fingers.
To translate the theoretical optimization into a practical, deployable solution, a physical prototype of the photovoltaic panel snow coverage monitoring and control device was developed. The hardware architecture of this device is anchored by an STM32F103C8T6 microcontroller, serving as the central processing unit. Sensor data acquisition is handled by a dedicated FDC2214 chip, specifically designed for precise capacitance measurements. An AM2320 sensor was integrated for accurate temperature and humidity monitoring, while an ESP8266 wireless communication module, interfaced with the ONENET cloud platform, provides seamless remote data transmission, enabling real-time monitoring from a PC terminal. The system is powered by a lithium battery, ensuring autonomous operation, and incorporates an OLED module for on-site data visualization. A comprehensive suite of functional tests was conducted. The results from these experiments confirmed the system’s operational effectiveness. The actual capacitance values measured by the device closely followed the theoretical trends predicted by the simulation model. The maximum relative error between the measured and simulated capacitance values was found to be 17.2%, which stabilized to approximately 13.4% after the system reached a steady state. These findings robustly demonstrate the practical utility of the designed monitoring system for accurately assessing snow accumulation on solar panels, thus providing a valuable and reliable tool for maintaining the efficiency and structural safety of solar energy installations in snowy climates.
1. Introduction
Solar power has become a critical part of the global energy mix, playing a pivotal role in ensuring clean, reliable, and sustainable electricity supply. The popularity of solar panels is due to their environmental benefits, decreasing costs, and versatility. However, the efficient operation of solar panels is highly dependent on environmental conditions. Among these, snow accumulation is a particularly challenging issue for solar panels installed in high-latitude or high-altitude regions. Snow covering the surface of solar panels blocks sunlight, leading to a significant reduction in power output. This not only decreases the energy yield but also poses structural risks to the panels due to the additional weight of heavy snow, potentially causing irreversible damage and substantial financial losses.
The core physical mechanism behind this issue lies in the optical and dielectric properties of the snow layer. Snow acts as a physical barrier, reflecting a large portion of incoming solar radiation and absorbing the rest, thereby preventing it from reaching the photovoltaic cells. For instance, a relatively thin layer of snow, only 20 mm thick, can reduce the solar radiation reaching the panels by up to 90%, catastrophically impacting energy production. The dielectric properties of snow, particularly its relative permittivity (dielectric constant), change dynamically with its density, water content, and physical state (e.g., dry snow, wet snow, or ice). This characteristic can be exploited for sensing purposes. An interdigitated capacitive sensor placed on the surface of solar panels can effectively detect these dielectric changes, providing a method to monitor the presence, thickness, and potentially the state of accumulated snow. This technology promises a more direct, localized, and accurate measurement compared to traditional weather monitoring or time-consuming manual inspection.
In light of these challenges, this research embarks on a comprehensive study with several key objectives:
- To design and optimize a sensor structure for effective snow detection on solar panels. This involves rigorous finite element analysis to understand the impact of various structural parameters.
- To develop a sophisticated multi-objective optimization methodology to fine-tune the sensor design for optimal performance.
- To design, build, and validate a complete monitoring system that not only detects snow but can also initiate a mitigation response, such as triggering a heating element for active snow removal from solar panels.
This thesis details the entire research process, from the theoretical underpinnings of capacitive sensing to the practical implementation and testing of a real-world monitoring device, contributing to the reliability and resilience of solar energy systems in winter climates.
2. Fundamental Principles of Interdigitated Capacitive Sensors
The operational principle of an interdigitated capacitive sensor is rooted in the way an electric field interacts with matter. To apply this technology effectively to snow monitoring on solar panels, a solid understanding of the dielectric properties of ice and snow is essential.
2.1 Dielectric Properties of Ice and Snow
The relative dielectric constant (εᵣ) is a key parameter that measures how a material responds to an electric field. It indicates how easily a material can be polarized by an electric field, which in turn affects the capacitance of a sensor placed in its proximity. The dielectric constant of snow is not a fixed value; it varies significantly depending on its composition, especially its volumetric water content (VWC). Dry snow is essentially a mixture of ice and air and has a relatively low dielectric constant (approximately 1.5 to 2). Wet snow, containing unfrozen water, exhibits a much higher dielectric constant because water is a highly polar molecule. This relationship, at a temperature of 0°C, can be modeled mathematically.
Based on data shown in a graph illustrating snow’s dielectric behavior, the following empirical model describes the relationship between the relative permittivity of snow (ε_snow) and its volumetric water content (VWC).
$$ \varepsilon_{snow} = 1.9 + 0.3905 \times (1 – e^{-VWC / 1.5}) \quad (2-1) $$
In this formula, a base value of 1.9 represents the dielectric constant of dry snow. The dynamic and sensitive nature of snow’s dielectric constant, which changes with its water content and density, provides a solid physical foundation for using capacitive sensors to monitor snow conditions on solar panels.
2.2 Theory of a Conventional Parallel-Plate Capacitor
The most fundamental type of capacitor is the parallel-plate capacitor, which consists of two conductive plates separated by an insulating material, or dielectric. Under an applied voltage, an electric field is created between the plates, storing electrical energy. For an ideal parallel-plate capacitor, where the edge effects are negligible, the capacitance (C) can be calculated using the following well-known formula:
$$ C = \frac{\varepsilon_b S}{4 \pi K d} \quad (2-2) $$
Here, C is the capacitance in Farads, ε_b is the relative permittivity of the dielectric material between the plates, S is the overlapping area of the plates in square meters, d is the distance between the plates, and K is the Coulomb’s law constant. This formula shows a direct proportional relationship between capacitance and the dielectric constant of the material filling the space between the plates. Therefore, if the area S and plate separation d are held constant, any change in the capacitance C can be attributed directly to a change in the dielectric material, ε_b.
However, in reality, a capacitor also has a fringe field that extends beyond the edges of the plates. Although often neglected in simple models, this edge effect can be significant, especially when the plates are thin and closely spaced. An approximate expression for the capacitance including edge effects can be given by a more complex equation, which highlights that the total capacitance is a sum of the main parallel-plate capacitance and a term related to the geometry of the edges. This fringe field is a core concept that is intentionally exploited in interdigitated capacitive sensors.
2.3 Structure and Working Principle of Interdigital Capacitive Sensors
An interdigitated capacitive sensor (IDC) is a planar, open-faced structure, which is fundamentally different from a parallel-plate capacitor. Instead of two stacked plates, an IDC consists of a set of coplanar, comb-like metal electrodes that are alternately connected, forming two electrically distinct terminals: the excitation electrode and the sensing electrode. When a voltage is applied to the excitation electrode, a non-uniform, semi-elliptical electric field (the fringe field) is generated between the electrodes, arching over the surface of the sensor.
This sensor is designed so that the fringing electric field extends outwards from the surface, penetrating into the medium above it, as illustrated in the sensor structure. A protective layer is often placed to shield the electrodes and prevent environmental contamination.
When a target material, such as a layer of snow, is placed on the top surface of the sensor, it enters the region where the fringe field is active. This material changes the effective dielectric constant of the space between and around the electrodes, which in turn alters the sensor’s output capacitance. By continuously measuring the capacitance value, it is possible to detect the presence of the material and, to a certain extent, deduce its properties, such as thickness.
The total capacitance (C_all) of an interdigitated structure with N pairs of fingers can be expressed as:
$$ C_{all} = C_0 \times (2N – 1) \quad (2-3) $$
C_0 is the capacitance of one elementary unit cell formed by a single pair of adjacent fingers. This unit capacitance C_0 can be broken down into three main components:
$$ C_0 = C_R + C_g + C_x \quad (2-4) $$
- C_R: The parallel-plate capacitance generated directly between the two vertical side walls of adjacent fingers.
- C_g: The capacitance generated in the region of the fringe field above the sensor that interacts with the target material.
- C_x: The capacitance associated with the fringe field penetrating through the substrate material.
The parallel plate component (C_R) can be approximated using a formula similar to that of the parallel plate capacitor, based on the geometry of the fingers.
$$ C_R = \frac{\varepsilon_0 \varepsilon_1 L d}{m} \quad (2-5) $$
Here, ε₀ is the vacuum permittivity, ε₁ is the effective dielectric constant of the material in the gap (which could be a polymer layer), L is the length of the fingers, d is the thickness of the electrode, and m is the finger spacing.
The combination of C_g and C_x, which accounts for the crucial edge and fringe field effects, can be more conveniently calculated using the conformal mapping method. This leads to the following expression:
$$ C_g + C_x = \varepsilon_0 L \frac{K(u’)}{K(u)} (\varepsilon_2 + \varepsilon_3) \quad (2-6) $$
In this equation, u is the “metallization factor,” a dimensionless geometric parameter defined as the ratio of the finger width (w) to the spatial period (the sum of the finger width and spacing). The factor u’ is its derivative (1-u). K(u) is the complete elliptic integral of the first kind. ε₂ and ε₃ are the dielectric constants of the material above the sensor (e.g., snow) and the substrate, respectively. The metallization factor u is given by:
$$ u = \frac{w}{w + m} \quad (2-7) $$
These equations form the foundational mathematical model for understanding how the geometry of the interdigitated sensor and the dielectric properties of the environment, specifically a snow layer on solar panels, govern the sensor’s output signal.
3. Finite Element Modeling and Analysis of the Sensor
While analytical models provide a strong foundation, they often rely on simplifications. To accurately predict the performance of an interdigitated capacitive sensor, especially for complex boundary conditions like a snow layer, numerical methods such as finite element analysis are required. In this research, COMSOL Multiphysics was used to construct a detailed and realistic three-dimensional model of the sensor for extensive simulation studies.
3.1 Model Construction
The simulation model was built as a 3D stack of four main layers: an air domain at the top, a snow layer underneath, the interdigitated capacitive electrodes, and a substrate at the bottom. The material properties were assigned as follows: snow was modeled with a relative permittivity of 3.5 and a thickness of 20 mm, while the air domain had a permittivity of 1. The sensor electrodes were made of copper, and the substrate was chosen as FR4, a common material for PCBs.
The geometry of the model was carefully defined, allowing for the parameterization of key structural attributes: finger length (L), finger width (w), finger spacing (m), and the total number of finger pairs (N). For the physics setup, the “Electrostatics” interface was used. A DC bias voltage of 5V was applied to one set of interdigitated electrodes (excitation), while the other set was grounded (sensing). The model was then discretized using free tetrahedral mesh elements. A finer mesh was applied to the delicate finger structure of the interdigitated capacitor to capture the rapidly varying field gradients and ensure accurate results, while a coarser mesh was used for the surrounding air domain to save computation time.
One of the primary outputs of this simulation is the visualization of the electric field distribution between the electrodes.
3.2 Sensor Performance Metrics
To systematically evaluate the performance of the different sensor designs, three key qualitative and quantitative indicators were defined:
- Signal Strength (Output Capacitance): This is the absolute value of the capacitance measured by the sensor. It represents the raw signal available for the measurement hardware and indicates the overall coupling efficiency of the sensor.
- Sensitivity (Sen): This is defined as the rate of change in output capacitance with respect to a change in the measurand, which in this case is the thickness of the snow layer. It effectively measures how responsive the sensor is to changes in snow thickness.
$$ S_{en} = \frac{\Delta C}{\Delta H} \quad (3-1) $$
Here, ΔC is the change in capacitance and ΔH is the corresponding change in snow depth. A higher sensitivity means the sensor can more effectively discern small variations in snow thickness.
- Maximum Detection Distance: This refers to the maximum distance from the sensor surface at which it can still accurately sense a change in the dielectric environment. It is defined as the distance where the output capacitance reaches about 98% of its saturation value when a thick object is introduced. This determines the effective range of the sensor.
3.3 Single-Factor Simulation and Analysis
To understand the individual impact of each geometric variable, a single-factor analysis was performed. In each study, all parameters except for the one under investigation were held constant, and the output capacitance and sensitivity were calculated for varying snow thicknesses.
Effect of Metallization Factor (u): The metallization factor was varied (0.4, 0.5, 0.6, and 0.7) to study its influence. The simulation results showed that a higher metallization factor correlates with a higher initial baseline capacitance. For example, with u=0.7, the initial capacitance was around 35 pF, while with u=0.4, it was approximately 30 pF. This factor also influenced sensitivity, as the capacitance curve for u=0.7 showed a steeper initial rise with snow thickness. All curves eventually saturated for snow thicknesses exceeding about 5 mm, but the saturated capacitance value was higher for higher metallization factors.
Effect of Finger Spacing (m): The spacing between fingers was varied between 0.3 mm and 0.9 mm. This parameter had a profound effect on the sensor’s performance. Smaller spacings like m=0.3 mm resulted in a significantly higher initial and saturated capacitance (about 49 pF at saturation) compared to m=0.9 mm (about 27 pF). This is because smaller gaps create a stronger coupling between adjacent electrodes. The sensitivity was also higher for the smaller spacing design, demonstrating a superior ability to detect thin snow layers.
Effect of Finger Length (L): The length of the fingers was varied from 20 mm to 80 mm. The simulation results showed a direct and pronounced effect. Both the output capacitance and sensitivity increased significantly with longer fingers. For example, when the finger length was increased from 60 mm to 80 mm, there was a marked jump in the saturated capacitance value. This is expected, as longer fingers increase the effective area over which the electric field interacts with the snow.
Effect of Number of Finger Pairs (N): The number of finger pairs was set to 6, 8, 10, and 12. This parameter had a nearly linear effect on the sensor’s output. The initial capacitance showed a direct correlation, increasing from about 30 pF for N=6 to 62 pF for N=12. The dynamic range of the capacitance change due to the snow was also larger for sensors with more finger pairs. All of these effects are consistent with the theoretical understanding that more interdigitated fingers increase the total effective area for charge storage and field generation.
3.4 Orthogonal Experiment and Parameter Ranking
While single-factor experiments reveal the individual effects, they do not show which factors are dominant when all of them interact. To understand this, an orthogonal experimental design, specifically an L16(4⁴) array, was implemented. This systematic approach allows the identification of the primary and secondary factors with a minimal number of simulation runs. The four parameters were treated as factors, each with four levels, as defined in the table below.
| Level | Factor A: Metallization factor (u) | Factor B: Finger spacing (m) | Factor C: Finger length (L) | Factor D: Number of finger pairs (N) |
|---|---|---|---|---|
| 1 | 0.4 | 0.3 | 20 | 6 |
| 2 | 0.5 | 0.5 | 40 | 8 |
| 3 | 0.6 | 0.7 | 60 | 10 |
| 4 | 0.7 | 0.9 | 80 | 12 |
The simulation results were then processed using a mean-range analysis method. This involves calculating the average output capacitance for each level of each factor and then determining the range (R), which is the difference between the maximum and minimum average values for each factor. A larger range value indicates a stronger influence of that factor on the response variable. The range is computed as follows:
$$ R = \max( \bar{M}_1, \bar{M}_2, …) – \min( \bar{M}_1, \bar{M}_2, …) \quad (3-2) $$
The following tables summarize the mean-range analysis for the two most critical metrics: output capacitance and maximum detection distance.
| Metric | Primary and Secondary Ranking | |||
|---|---|---|---|---|
| Output Capacitance and Sensitivity | Finger Length (L) | > Number of Pairs (N) | > Finger Spacing (m) | > Metallization Factor (u) |
| Maximum Detection Distance | Metallization Factor (u) | > Finger Spacing (m) | > Number of Pairs (N) | > Finger Length (L) |
The orthogonal analysis revealed that for achieving a high signal strength (output capacitance) and sensitivity, the most critical design parameter to adjust is the finger length, followed closely by the number of finger pairs. This provides a clear and scientifically justified direction for prioritizing sensor design improvements for solar panels. However, this analysis does not directly address the trade-off with sensor size, which is crucial for practical deployment.
4. Multi-Objective Structural Parameter Optimization Using a Joint Optimization Algorithm
The performance of the sensor is dictated by a complicated interplay of structural parameters. Increasing the length and number of fingers enhances the signal strength but increases the physical footprint of the sensor, which is an undesirable trait for low-profile integration on solar panels. This leads to a classic multi-objective optimization problem: we aim to maximize the output capacitance while simultaneously minimizing the total sensor area. Due to the non-linear and complex nature of this relationship, an efficient global optimization strategy is required.
4.1 Methodology of the Joint Optimization
This research introduces a sophisticated hybrid optimization method combining several advanced algorithms. The process begins by building a highly accurate predictive model that can mimic the finite element simulations but with much lower computational cost. This is accomplished using a Least Squares Support Vector Machine (LSSVM). The performance of the LSSVM model is highly sensitive to its two user-defined hyperparameters: the regularization parameter (γ) and the radial basis function kernel bandwidth (σ). Manually tuning these parameters is tedious and inefficient. Therefore, an improved meta-heuristic algorithm called IGOOSE was developed and used for automatic hyperparameter optimization.
The IGOOSE Algorithm is an enhancement of the GOOSE (Goose Optimization Algorithm) and incorporates three key modifications to improve its search efficiency and avoid premature convergence:
- Piecewise Chaotic Mapping: Is used to initialize the population of “geese” in the search space. Chaotic sequences have better ergodicity and diversity than random sequences, ensuring that the initial population is spread more uniformly across the search domain, which can lead to a faster convergence.
- Golden Sine Strategy: Is integrated into the exploration phase. This strategy helps to update the position of the optimal solution by scanning the surrounding space in a more efficient way, guided by the golden ratio, thereby improving both local search capability and the ability to escape from local optima.
- Nonlinear Dynamic Adaptive Inertia Weight: Replaces a constant or linearly decreasing α parameter in the algorithm’s position update formula. This dynamic weight, which adjusts its value non-linearly based on the iteration progress and a random factor, allows the algorithm to prioritize global exploration in early iterations and shift towards local exploitation in later stages, achieving a better balance between the two.
The performance of the IGOOSE algorithm was rigorously benchmarked against the original GOOSE algorithm using a set of 19 standardized CEC2005 test functions. The results, shown in the table below, demonstrated that the IGOOSE algorithm generally achieved better optimal values, mean values, and standard deviations across a majority of the functions, confirming its enhanced global search ability and robustness.
| Index | Optimal Value | Mean Value | Standard Deviation | |||
|---|---|---|---|---|---|---|
| GOOSE | IGOOSE | GOOSE | IGOOSE | GOOSE | IGOOSE | |
| F1 | 4.10E-60 | 5.32E-140 | 466.5 | 25.08 | 4412.8 | 526.4 |
| F8 | -6640 | -9016.3 | -4278.9 | -8988.9 | 1272.3 | 261.1 |
| F14 | 2.98 | 0.998 | 3.006 | 1.0319 | 0.678 | 0.187 |
| F18 | 3 | 3 | 3.112 | 3.0000 | 2.179 | 0.0021 |
4.2 Predictive Model Construction and Validation
To generate a robust dataset for training and validating the LSSVM model, the Latin Hypercube Sampling (LHS) technique was employed. LHS is a statistical method for generating a near-random sample of parameter values from a multidimensional distribution. It is an efficient way to cover a large parameter space uniformly. A total of 240 sample points were generated, each representing a unique combination of the four structural parameters (u, m, L, N). These parameter sets were used in the COMSOL model to calculate the corresponding output capacitance and physical size. The ranges for the parameters were defined to cover the practical range of interest for the sensor design.
| Parameter | Symbol | Range |
|---|---|---|
| Metallization factor | u | 0.3 ~ 0.7 |
| Interdigital spacing (mm) | m | 0.3 ~ 0.9 |
| Finger length (mm) | L | 40 ~ 50 |
| Finger pairs | N | 6 ~ 10 |
The dataset was split into a training set (204 samples) and a testing set (36 samples). The IGOOSE-optimized LSSVM model was then trained on this data. Its performance was compared to other benchmark models like PSO-LSSVM, WOA-LSSVM, and GOOSE-LSSVM.
| Model | For Capacitance Prediction | ||
|---|---|---|---|
| MAE | MAPE | R² | |
| PSO-LSSVM | 1.8327 | 0.0359 | 0.9144 |
| WOA-LSSVM | 1.5367 | 0.0286 | 0.9458 |
| GOOSE-LSSVM | 1.3141 | 0.0263 | 0.9746 |
| IGOOSE-LSSVM | 0.6756 | 0.0126 | 0.9951 |
| Model | For Size Prediction | ||
|---|---|---|---|
| MAE | MAPE | R² | |
| PSO-LSSVM | 25.5696 | 0.0253 | 0.9603 |
| WOA-LSSVM | 17.2977 | 0.0150 | 0.9777 |
| GOOSE-LSSVM | 26.1658 | 0.0285 | 0.9721 |
| IGOOSE-LSSVM | 20.9767 | 0.0245 | 0.9956 |
The evaluation metrics clearly show the superiority of the proposed IGOOSE-LSSVM model. The near-perfect R² values (0.9951 for capacitance and 0.9956 for size) indicate that the prediction model can substitute the finite element simulations with a high degree of accuracy, making the subsequent multi-objective optimization computationally feasible.
4.3 Multi-Objective Optimization and Results
The final phase of the optimization is to use the IGOOSE-LSSVM predictive model as the fitness function within the NSWOA algorithm. The goal of this multi-objective optimization is to solve for a set of Pareto-optimal solutions that balances the two conflicting objectives. The mathematical framework of this optimization problem is formulated as follows:
$$ \begin{aligned} \text{Minimize / Maximize} \quad & F(X) = (u, m, L, N) \\ \text{Subject to:} \quad & X_{min} \leq X \leq X_{max} \\ \text{Objectives:} \quad & \text{Minimize } A = A(X) \\ \text{} \quad & \text{Maximize } C = C(X) \end{aligned} \quad (4-1) $$
Where $X$ is the vector of design variables, $A(X)$ is the total sensor size, and $C(X)$ is the output capacitance. The algorithm seeks to find a set of non-dominated solutions, meaning solutions where improving one objective (e.g., increasing capacitance) inevitably worsens the other objective (e.g., increasing size). The resulting Pareto front provides a set of optimal design choices for the engineer. From the resulting set of optimal solutions shown in the Pareto front chart, a solution that offers a balanced trade-off between size and capacitance was selected. To ensure the parameters comply with practical manufacturing capabilities, the optimal values were rounded. The final optimized design parameters are presented below.
| Parameter | Metallization Factor (u) | Spacing (m) | Length (L) | Finger Pairs (N) |
|---|---|---|---|---|
| Optimized Values | 0.48 | 0.33 mm | 47.46 mm | 9 |
The performance of the optimized sensor was then verified against a baseline design. The results of this comparison are summarized in the table below. The joint optimization approach yielded a design with a significantly reduced footprint and an increased capacitance output, successfully overcoming the trade-off between size and performance.
| Performance Metric | Baseline | Optimized | Improvement |
|---|---|---|---|
| Peak Capacitance | – | – | +35.61% |
| Peak Sensitivity | 28.5 pF/mm | 31.9 pF/mm | +11.9% |
| Sensor Area | 787.2 mm² | 526.5 mm² | -33.1% |
5. Design and Implementation of the Monitoring Device
Following the successful theoretical optimization of the sensor, this chapter focused on translating this design into a functional, real-world monitoring and control device. The objective was to create a stand-alone unit that could be mounted on a photovoltaic panel, continuously measure environmental conditions, and actively respond to snow accumulation by initiating a heating system.
5.1 Hardware Design
The hardware architecture of the device was designed with a modular approach for ease of assembly and maintenance. A schematic overview of this architecture was presented. The principal components are:
- Main Control Unit: An STM32F103C8T6 microcontroller is at the heart of the device. It orchestrates all functions, including reading sensor data, processing information, controlling display, and managing communication.
- Power Supply Module: A 3.7V lithium-polymer battery supplies power, enabling true remote and autonomous operation. A USB interface is included for recharging and for initial programming.
- Sensing Element: The custom-designed interdigitated capacitive sensor, based on the optimized parameters, is mounted externally to interface directly with the snow layer.
- Capacitance Measurement Module: The FDC2214 chip is used for high-precision capacitance measurements. It works based on the principle of LC resonance; the sensor is part of an LC tank circuit, and changes in the sensor’s capacitance shift the resonant frequency, which the chip detects and converts into a digital value. The chip communicates with the MCU via the IIC protocol. A 40MHz crystal oscillator was added for improved measurement stability.
- Temperature and Humidity Sensor: An AM2320 sensor was chosen to monitor ambient conditions. It uses an IIC interface and can measure temperature (-40°C to 80°C) and humidity (0-99.9% RH) with high accuracy.
- Display Module: An OLED display with a 128×64 resolution was selected to provide real-time local readouts of sensor data and system status.
- Wireless Communication Module: An ESP8266 Wi-Fi chip is used to transmit monitoring data to the cloud. It connects to a local Wi-Fi network and uses the MQTT protocol to send data to the ONENET cloud platform, enabling remote access from anywhere.
- Snow Removal Module: For active snow removal, a heating element controlled by a 5V relay is integrated. When the capacitance signal indicates that a critical snow thickness threshold has been exceeded, the microcontroller activates the relay, which supplies power to the heater, melting the snow and preventing hazardous accumulation on the solar panels.
5.2 Software Design
The software for the embedded system was developed in the Keil5 integrated development environment using the C language. The system flow is designed for efficiency and remote operation, adhering to a cycle of data collection, transmission, and power management.
The main program flow is as follows:
- System Initialization: Upon power-up, the MCU initializes its system clock, I/O ports, communication interfaces (IIC, UART), and the watchdog timer.
- Data Acquisition: The system enters a low-power monitoring state. It sends a command to wake up the AM2320 sensor and reads the temperature and humidity values. Simultaneously, it reads the capacitance value from the FDC2214 via the IIC interface.
- Data Processing and Upload: The collected data is formatted and sent to the ESP8266 module via a serial port. The ESP8266, in turn, uses MQTT to publish the data to the ONENET cloud platform. This allows for near-real-time visualization on a PC.
- Control Logic: The MCU compares the recorded capacitance value, which is directly related to snow depth, with a pre-set alarm threshold. If the capacitance exceeds this threshold, which indicates a significant snow layer on the sensor (and implicitly on the surrounding solar panels), the MCU triggers the relay to turn on the heating pad.
- Power Management: To preserve battery life, the MCU and the ESP8266 enter deep sleep modes after completing a transmission cycle. A periodic wake-up timer is used to initiate the next data collection cycle. The system could also be configured to remain in a low-power state until an interrupt (e.g., from a simple threshold comparator) is triggered.
The PC-side software visualizes the data by creating a user-friendly interface that logs into the ONENET cloud platform, retrieves real-time data streams, and displays them on a dashboard. This allows the user to monitor the operational status of the solar panels, view historical data trends, and receive alert notifications, all from a remote location.
5.3 Enclosure Design
A durable enclosure was designed and fabricated to protect the sensitive electronics in outdoor environments. The design focused on three main aspects:
- Conformability: The internal layout was designed to securely fit the main control board and battery, minimizing empty space.
- Sealing: To prevent moisture and snow from damaging internal components, the enclosure was designed to be watertight and airproof, creating an isolated internal environment.
- Corrosion Resistance: The material was chosen for its ability to withstand UV radiation, extreme temperature fluctuations, and the corrosive effects of prolonged exposure to moisture and snow.
6. System Testing and Analysis
This chapter details the experimental evaluation of the fabricated solar panel snow monitoring system. The primary goal was to validate the functionality of the entire system and assess the accuracy of its snow detection capabilities in real-world scenarios.
6.1 System Functionality Tests
Before deploying the system for field testing, several basic functional checks were conducted.
- Communication Test: The connection between the STM32 MCU and the ESP8266 Wi-Fi module was tested, and the successful transmission of data packets to the ONENET cloud was verified. The clickable webpage was successfully accessed, confirming the remote data relay functionality.
- Cloud Platform Interaction Test: The PC interface was accessed, and real-time data from the sensors was successfully retrieved and displayed on the dashboard. This verified data logging and visualization. The real-time monitoring interface successfully displayed four data points for capacitance and a single value for temperature and humidity.
- Capacitance Data Acquisition Test: The sensor was tested by introducing objects into its sensing field, and the output capacitance was observed to change accordingly on the OLED screen. For example, the capacitance values of channels 1-4 were measured as 36 pF, 39 pF, 41 pF, and 44 pF, respectively, when different targets were placed near the sensor. This confirmed that the entire signal chain, from the sensor to the display, was working correctly.
6.2 Real Snow Monitoring Test and Comparative Analysis
To evaluate the practical detection accuracy of the system, a snow monitoring experiment was performed. The sensor was fixed onto the surface of a solar panel, and a controlled amount of snow was applied to create a snow layer of up to 10 mm in thickness. The sensor’s capacitance was continuously recorded and compared with the values predicted by the finite element simulation model. The test results are shown in table format below.
| Snow Thickness (mm) | Simulated Capacitance (pF) | Actual Capacitance (pF) | Relative Error (%) |
|---|---|---|---|
| 2 | 65.5 | 76.7 | 17.2% |
| 4 | 81.2 | 88.8 | 9.3% |
| 6 | 88.3 | 98.2 | 11.2% |
| 8 | 91.4 | 103.4 | 13.4% |
| 10 | 92.8 | 105.1 | 13.4% |
As illustrated in the table, the capacitance values from both the simulation and the actual sensor followed a very similar trend. Both curves show a rapid initial increase in capacitance as the snow thickness increases from 0 to about 4 mm, followed by a gradual plateau and saturation as the snow layer gets thicker. This rapid response in the early stages confirms the sensor’s high sensitivity to thin snow layers, which is critical for early warning.
However, the measured capacitance values were consistently higher than the simulated values across the entire range. The relative error was at its highest (17.2%) when the snow layer was very thin (1 mm), decreased to a minimum of 9.3% at a moderate thickness of 3 mm, and then stabilized at around 13.4% for thicker snow layers.
The observed discrepancies between the simulated and actual measured values can be attributed to several factors:
- Idealized Properties of Snow: The simulation model assumes a perfectly homogenous snow layer with uniform density and a fixed dielectric constant. In reality, natural snow is a complex mixture of ice, air, and sometimes liquid water, and its properties vary significantly from test to test and even within the same sample. Variations in snow density and water content directly affect its overall dielectric constant, which causes the discrepancy.
- Sensor-Snow Interface: The simulation assumes perfect contact between the sensor and the snow. In actual conditions, there are microscopic air gaps at the interface, and the physical contact may not be uniform. These micro-gaps alter the electric field distribution and reduce the measured capacitance.
- Environmental Factors: During the actual experiment, ambient conditions like temperature, humidity, and solar radiation can influence the snow’s state and dielectric properties, causing deviations from the static conditions assumed in the simulation.
7. Conclusion and Future Work
This thesis has successfully navigated the complete lifecycle of a smart sensor system, from theoretical modeling and advanced optimization to practical implementation and testing, specifically tailored for monitoring snow on solar panels. The key contributions and findings of this research are summarized below.
First, the research established a comprehensive simulation framework. Using COMSOL finite element analysis, a detailed 3D model of an interdigitated capacitive sensor was built. Orthogonal experiments and range analysis identified the critical design parameters and their relative importance for sensor performance. It was found that for maximizing output capacitance and sensitivity, the finger length and number of finger pairs are the dominant factors.
Second, a novel and powerful optimization methodology, IGOOSE-LSSVM-NSWOA, was developed and validated. This multi-faceted approach successfully balanced the conflicting goals of high signal strength and small physical size. The IGOOSE algorithm proved superior to its predecessor, and the IGOOSE-LSSVM surrogate model accurately predicted sensor behavior, dramatically reducing computational cost. The multi-objective optimization yielded an optimal sensor design with a parameter set of a metallization factor of 0.48, spacing of 0.33 mm, length of 47.46 mm, and 9 finger pairs. This optimal design resulted in a 35.61% increase in capacitance, an 11.9% increase in sensitivity, and a 33.1% reduction in size compared to a baseline design.
Third, a fully functional prototype was constructed and successfully tested in real-world conditions. The device, built around an STM32 microcontroller, integrates the optimized sensor, and communication modules for remote monitoring of solar panels. The field tests validated the system’s ability to monitor snow coverage effectively. The strong agreement between the measured and simulated capacitance values confirmed the accuracy of the modeling and the engineering practicality of the entire monitoring system. Although minor discrepancies existed, they were well-within acceptable limits for practical application and are attributable to the highly variable nature of natural snow.
Looking towards the future, several avenues of research can further enhance this system:
- Enhanced Snow Type Recognition: The current sensor demonstrates high effectiveness in detecting snow on solar panels. Future research could explore the use of multi-frequency impedance spectroscopy to not only detect snow but also differentiate between snow types, such as dry snow, wet snow, and ice. This would provide more detailed information about the state of the snow, which is valuable for refining the de-icing strategy and improving the accuracy of power generation forecasts.
- Power Optimization: Further work could be directed towards improving the energy autonomy of the system. This could involve implementing more sophisticated solar-powered energy harvesting to recharge the battery, ensuring indefinite, maintenance-free operation. Optimizing the sensor array to be active only when a simple threshold is crossed could also significantly reduce power consumption.
- Scalability and Wireless Sensor Networks: The proposed device was designed as a single unit. Future research could investigate a network topology where a master unit can communicate with multiple slave sensor nodes placed on different sections of a solar panel array. This would allow for intelligent, site-specific snow removal strategies, optimizing energy usage across the whole photovoltaic installation.
The system developed in this work provides a valuable tool for the efficient and safe operation of solar panels in wintery climates, contributing to the global transition towards more resilient and reliable renewable energy sources.