Flexible Wireless Vibration Sensing for Table Grape in Cold Chain

Abstract

The quality change process of table grapes during cold chain logistics is complex and highly susceptible to vibration-induced damage. Traditional monitoring techniques not only consume significant human and material resources but also cause destructive effects on the fruit structure of table grapes, making them difficult to apply in practical scenarios. Based on this, this paper focuses on table grapes in cold chain business processes and designs a flexible wireless vibration sensor for monitoring the quality of table grapes during cold chain transportation. The hardware component of the system fabricates a flexible wireless vibration sensing for monitoring the quality of the table grape cold chain. In contrast, the software component develops corresponding data acquisition and processing functionalities. Using Summer Black table grapes purchased from Tianjin Hongqi Agricultural Market as the research subject, correlation and quality monitoring models for the cold chain process of table grapes were constructed. After Z-score standardization, the prediction results based on the MLR model achieved R² values all greater than 0.87 and RPD values all exceeding 2.7. Comparisons with other regression models demonstrated its optimal fitting performance for monitoring the quality of the cold chain for table grapes. This achieves non-destructive and high-precision data acquisition and processing during the cold chain process of table grapes, wirelessly transmitting results to terminal devices for real-time visual monitoring.

1. Introduction

China, as the country with the largest cultivation area of table grapes, produced 14.2 million tons of table grapes in 2024, with continuously growing export volumes. As living standards improve, consumers increasingly demand higher-quality and fresher fruit. They no longer settle for purchasing local, seasonal table grapes but expect to enjoy fresh grapes with excellent taste and quality, even outside grape-producing regions and out of season [1,2,3].

Table grapes, characterized by high water content, thin skin, and tender texture, are prone to decay due to microbial infection and physiological changes [4]. Their shelf life at room temperature is only a few days; prolonged storage leads to mold, softening, and rotting, which limits their sales scope and duration [5]. After harvest, table grapes exhibit vigorous respiration, continuously consuming nutrients and producing heat, carbon dioxide, ethylene, and other metabolites [6]. This process decomposes sugars, reducing sweetness, deteriorating flavor, accelerating ripening and senescence, and shortening the shelf life [7,8,9]. Consequently, the cold chain for table grapes has emerged, extending shelf life through refrigeration technology to expand the sales scope and duration, while ensuring fruit quality [10].

During cold chain transportation, vibration sources arise from vehicle movement, jostling during loading or unloading, and operation of refrigeration equipment [11]. Under vibration, grape berries collide with each other. Due to their delicate skin, this easily causes abrasions that may compromise skin integrity and accelerate spoilage [12]. Prolonged vibration may also cause packaging box breakage or looseness of the seal. Damaged packaging loses its protective function, making the berries more vulnerable to physical impact and compression, which further increases the risk of damage [13]. Additionally, compromised packaging alters internal gas composition and humidity, potentially causing surface condensation or dehydration, which heightens microbial growth risks [14]. Moreover, vibration enhances the respiration of table grapes, increasing sugar consumption and carbon dioxide production, which negatively affects flavor [15]. It also induces higher ethylene production, significantly accelerating ripening, shortening shelf life, and reducing commercial value [16].

The challenges in cold chain transportation of table grapes highlight the critical need for equipment capable of real-time, non-destructive monitoring of vibration, temperature, and humidity [17]. Traditional quality monitoring methods often struggle to meet the demands of modern industry. Manual sensory inspections—assessing color, size, shape, and empirically judging sweetness, acidity, etc.—are highly subjective, inconsistent, inefficient, and lack real-time capability [18]. These methods primarily focus on appearance and simple physical indicators, failing to accurately evaluate internal quality metrics such as sweetness, acidity, and maturity [19]. Continuous monitoring throughout transportation and storage is unfeasible, hindering data recording, traceability, and integrated analysis [20]. This impedes quality control improvements and precise issue tracing, resulting in significant lag. Furthermore, traditional rigid and destructive monitoring methods are incompatible with grape morphology, potentially damaging tender berries and causing inaccurate data or blind spots [21,22,23].

In summary, designing a flexible wireless vibration sensor for table grape cold chain quality monitoring is of significant importance and has considerable application value. This paper proposes a flexible wireless vibration sensing dedicated to monitoring quality changes in table grapes during cold chain processes. The system offers unique advantages: fabricated with flexible materials, it exhibits excellent flexibility and bendability, enabling close conformity to irregular grape surfaces or packaging. This ensures complete sensor contact with the monitoring target, achieving rapid perception, non-destructive detection, real-time tracking, and high-precision assessment of cold chain quality. Precise monitoring effectively reduces food safety risks caused by quality deterioration and minimizes unnecessary waste. Additionally, this sensing accumulates valuable experience in applying vibration technology for the quality monitoring of fruits and vegetables, promoting its further development and widespread adoption in the industry, thereby safeguarding the quality of agricultural products and advancing agrarian modernization.

This study focuses on table grapes in cold chain logistics, analyzing the damage mechanisms of quality changes and the correlation between vibration sensing data and quality indicators (berry drop rate, flesh firmness, skin firmness, pedicel pull force). Data is acquired via flexible wireless vibration sensing and processed using chemometrics to establish relevant quality monitoring models. Employing refined fabrication techniques, a flexible wireless vibration sensing is designed for monitoring the quality of the table grape cold chain, ensuring the stable operation of all components.

2. Materials and Methods

2.1. Overall Design

The cold chain business process for table grapes is illustrated in Figure 1a, primarily including grape harvesting, packaging, refrigeration, transportation, and sales [24]. Among these, the transportation segment faces significant challenges. During cold chain transportation, vibration sources arise from vehicle movement, jostling during loading or unloading, and operation of refrigeration equipment. Vibration can degrade the appearance and commercial value of table grapes [25]. Given the high correlation between vibration conditions and table grape quality during cold chain transportation, this study employs wireless sensor network technology as the monitoring method, designing and developing a flexible wireless vibration sensing for monitoring table grape cold chain quality. This sensing achieves non-destructive, high-precision, real-time dynamic monitoring throughout the cold chain process.

The architectural design diagram for the wireless data acquisition system for table grape cold chain quality monitoring is shown in Figure 1b. Architecturally, nodes in the wireless sensor network collect temperature, humidity, vibration, and other data during cold chain transportation and storage, transmitting them via GPRS or other means. The gateway aggregates and converts data, performing preliminary processing before transferring it to the service center via the Internet. Firewalls provide security by preventing unauthorized external access and cyberattacks, ensuring data safety [26]. Within the service center, application servers run data analysis programs for in-depth processing of received data. The model database stores predefined analytical models for assessing grape quality status, while the central database archives historical and real-time data to support quality traceability and analysis. Managers can access the cloud platform via laptops or other devices over the Internet, enabling real-time monitoring of the table grape cold chain status. This allows the timely detection of anomalies and intervention to maintain stable quality throughout the cold chain process.

The functional architecture design diagram of the flexible wireless vibration sensor is presented in Figure 1c. The microcontroller ESP8266 controls the sensing, which comprises the HDC1080 temperature and humidity sensor and the BMA220 accelerometer, to collect data. The Arduino platform implements the sensing functionality. Subsequently, the collected sensor data is processed, and the integrated WiFi module enables wireless data transmission to host computers and other terminal devices via WiFi connection using the MQTT protocol, as depicted in Figure 1d,e.

2.2. Flexible Wireless Vibration Sensing Preparation

Flexible circuits can conform well to the packaging surfaces of damage-prone table grapes, demonstrating excellent applicability in the cold chain [27]. This vibration sensing employs a flexible substrate as the PCB, specifically a flexible printed circuit board (FPC). Based on the designed hardware circuit schematic shown in Figure 2a, the printed circuit board (PCB) layout was created using double-sided routing, which significantly expands wiring space and flexibility, as illustrated in Figure 2b.

According to design requirements, the FPC production process begins with drilling: flexible copper-clad laminates (typically composed of copper foil, polyimide, and coverlay) are cut to appropriate sizes and shapes, and drilling machines are used to create holes for connecting circuit layers. The second step is the copper deposition process, aimed at depositing a thin layer of copper on the hole walls to make them conductive, as shown in Figure 2c, enabling electrical connections between different circuit layers. The third step is the circuit patterning process: a photosensitive dry film is evenly coated on the surface of the copper-clad laminate. Through exposure, development, and other steps, the circuit pattern on the dry film is transferred onto the copper foil surface. During exposure, ultraviolet light is shone through a circuit film onto the dry film, causing the exposed areas to undergo a photochemical reaction and become insoluble in the developer. The unexposed areas, however, can be dissolved by the developer, revealing the copper foil portions that need to be etched, as illustrated in Figure 2d. The final pattern plating process involves electroplating the surface of the already etched circuits: the board is placed into an electroplating tank, and using electroplating equipment, an electric current is passed through the electrolyte to deposit an additional layer of copper on the circuit surface. This further thickens the copper layer, enhancing the circuit’s conductivity and wear resistance. Depending on requirements, other metals such as tin may also be plated to meet different electrical performance and reliability standards.

After completing these four manufacturing steps, the structure of the flexible wireless vibration sensor begins to take shape. An exploded structural view and the substrate materials are presented in Figure 2e. Figure 2f shows the original physical image of the flexible circuit, clearly displaying its initial structure and appearance. Measuring 5.2 mm × 4 mm with a board thickness of 0.11 mm, it is highly lightweight and exhibits excellent flexibility.

The sensing hardware comprises a master control module, a sensing module, and a wireless transmission module, with the overall design block diagram shown in Figure 1c. The hardware circuit schematic of the master control module is presented in Figure 3a. The ESP8266 chip serves as the core, integrating full microcontroller functionality. Equipped with Wi-Fi capability, the ESP8266 operates in STA mode. In this sensing’s cold chain quality monitoring application, STA mode enables the ESP8266 to connect to networks and transmit sensor-collected data to remote servers, as shown in Figure 3b. The hardware circuit schematic of the sensing module is shown in Figure 3c. The sensing module consists of a vibration sensing submodule and a temperature-humidity sensing submodule. The core of the vibration sensing submodule is the BMA220 accelerometer chip from Bosch, with a compact package size of only 2 mm × 2 mm × 0.98 mm. Internal chip details are provided in Figure 3d. Its selectable acceleration measurement ranges (±2 g, ±4 g, ±8 g, ±16 g) meet the precision and range requirements for vibration measurement in cold chain scenarios. The core of the temperature-humidity sensing submodule is the HDC1080 chip, a low-power, high-precision digital temperature and humidity sensor. A typical application diagram for the HDC1080 is shown in Figure 3e, illustrating its connection to an external microcontroller via an I²C interface.

2.3. Experimental Scheme

This experiment aims to simulate the transportation process of table grapes under different vibration conditions. By measuring changes in the physiological indicators of table grapes, it delves into the relationship between vibration data and quality deterioration. The overall test plan is illustrated in Figure 4a. The table grapes (Summer Black variety) were purchased in August from Tianjin Hongqi Agricultural Wholesale Market, having been harvested and delivered to the market on the same day. Several clusters of grapes were selected based on consistent variety, uniform maturity, the absence of obvious pests/diseases, and no mechanical damage, all of which exhibited excellent quality.

The test plan simulated the cold storage environment of table grape cold chain logistics. Figure 4b illustrates the external cold chain environment in which the sensing was placed. The ambient temperature was precisely controlled at 1 °C, and humidity was maintained at 30%. Vibration was applied using a SHA-C reciprocating constant temperature oscillator (specifications: 49 × 39 × 22 cm). The vibration test chamber was secured to the oscillator using transparent adhesive tape. To prevent random movement, the grapes were arranged in a tightly packed arrangement inside the chamber. The flexible wireless vibration sensor was placed within the test chamber and connected to a small battery pack for power supply. The chamber was then sealed rigorously. As shown in Figure 4c, the oscillator employs stepless speed regulation, allowing flexible and precise adjustment of vibration speed to meet varying test requirements.

This study considered the diverse vibration scenarios that may occur during actual cold chain transportation and designed six different vibration treatment protocols. The vibration duration was uniformly set to 3 h, with vibration data accurately collected at 10-s intervals per cycle. For each vibration treatment protocol, 40 independent experiments were strictly repeated to effectively minimize experimental errors through multiple trials. As shown in Figure 4d, a vibration detector was installed to precisely validate the vibration data.

(1)

Cold-chain quality data acquisition

After completing the vibration data acquisition experiment for table grape cold chain quality, key indicators, including berry drop rate, flesh firmness, skin firmness, and pedicle pull force, were measured to accurately assess quality changes in the cold chain environment. The specific methods are as follows:

(a)

Measurement of Berry Drop Rate

The entire box of vibration-treated table grapes was placed on a clean electronic scale. The tray was gently shaken to separate naturally detached berries from the pedicles. The berry drop rate is calculated as:

$$L={\displaystyle \frac{M_2}{M_1}}\times 100\%$$

(1)

where M₁ represents the total mass of the entire box of grapes before the experiment, and M₂ represents the total mass of all detached berries after gently shaking the tray.

(b)

Measurement of Flesh Firmness

A disk probe was securely installed on the force measurement device of the texture analyzer. The measurement mode was set to the Texture Profile Analysis (TPA) mode. The disk probe’s operating speed was precisely set to 2.5 mm/s, the post-test speed to 5 mm/s, the trigger force to 5 g, the interval between two compressions to 5 s, and the compression degree to 20%.

(c)

Measurement of Pedicle Pull Force

Using scissors, several individual berries with intact pedicles and intact pedicle-berry junctions were randomly selected from the grape samples. The pedicel end of the berry was fixed to the upper fixture of the tensile testing machine, allowing the berry to hang freely. The berry was then pulled downward rapidly until separation occurred at the pedicel-berry junction. The maximum force recorded at rupture was noted. This process was repeated multiple times, and the average value was taken as the pedicle pull force.

(d)

Measurement of Skin Firmness

For measuring the skin firmness of table grapes, a needle probe was selected. The operating speed of the needle probe was set to 2.5 mm/s. The probe was retracted immediately after puncturing the grape skin. The post-test speed was set to 5 mm/s, and the trigger force was 5 g. This mode enables precise measurement of the rupture strength of the table grape skin.

(2)

Correlation model between vibration data and cold-chain quality damage

During the cold chain transportation of table grapes, vibration may cause quality deterioration. Thoroughly analyzing the correlation between vibration data and quality deterioration is crucial for optimizing grape cold chain logistics, reducing losses, and enhancing economic benefits. Theoretical analysis provides valuable references for research in this field.

First, data collection and dataset processing were performed. The Pearson correlation coefficient was used to measure the correlation between vibration data and the deterioration of cold chain quality. The Pearson correlation coefficient quantifies the linear relationship between two continuous variables, ranging from −1 to 1. The calculation method for the population correlation coefficient is shown in Formula (2):

$${\rho }_{X,Y}={\displaystyle \frac{\mathrm{Cov}(X,Y)}{{\sigma }_X{\sigma }_Y}}={\displaystyle \frac{E((X-{\mu }_X)(Y-{\mu }_Y))}{{\sigma }_X{\sigma }_Y}}$$

(2)

where Cov(X, Y) is the covariance of variables X and Y; σ_x and σ_γ are the standard deviations of X and Y, respectively; E denotes the expectation; and μ_x and μ_γ represent the means of X and Y.

The calculation method for the sample correlation coefficient is shown in Formula (3):

$$r={\displaystyle \frac{{\displaystyle \sum _{i=1}^n}(X_i-\overline{X})(Y_i-\overline{Y})}{\sqrt{{\displaystyle \sum _{i=1}^n}{(X_i-\overline{X})}^2\cdot {\displaystyle \sum _{i=1}^n}{(Y_i-\overline{Y})}^2}}}$$

(3)

where Xi and Yi are the i-th observed values of variables X and Y, $\overline{X}$ and $\overline{Y}$ represent the sample means of X and Y, and n is the sample size.

A correlation coefficient close to 1 indicates a strong positive linear relationship between the two variables, while a coefficient close to −1 indicates a strong negative linear relationship. A coefficient close to 0 suggests no significant linear correlation. Finally, all evaluation metrics were calculated and output, and relevant charts were generated.

(3)

Cold-chain quality monitoring model

Modeling for monitoring the quality of table grapes during cold chain logistics is a crucial means to ensure quality stability and reduce losses. By establishing a Multiple Linear Regression (MLR) model, the relationship between vibration data and key parameters characterizing the quality of table grapes in the cold chain—such as shattering rate, flesh firmness, skin firmness, and pedicel tensile strength—is investigated. MLR typically refers to the Multiple Linear Regression model, a classical statistical analysis method used to study the linear relationship between multiple independent variables (explanatory variables) and one dependent variable (response variable). This model is based on fundamental assumptions such as linearity, independence, homoscedasticity, and normality. It employs the least squares method to estimate parameters and uses metrics like the coefficient of determination (R²) and correlation tests for model evaluation. The goodness of fit is measured by R², where a value closer to 1 indicates a better fit. This approach enables the prediction of quality changes in table grapes during transportation and storage in cold chain environments, allowing for early understanding of their quality status under different vibration conditions.

Step 1. Collect cold chain environmental data and quality indicator data for table grapes to establish the dataset.

Step 2. Perform data preprocessing. Apply Z-score standardization to the collected data, removing outliers and handling missing values.

Step 3. Split the dataset into training and testing sets at an 8:2 ratio. The training set is used to train the model, enabling it to learn patterns and relationships within the data; the testing set is used to evaluate the generalization capability of the trained model on unseen data, i.e., the model’s prediction accuracy for new data.

Step 4. Establish the table grape cold chain quality monitoring model. Develop an MLR (Multiple Linear Regression) model to analyze the linear relationships between multiple environmental factors and table grape quality indicators, predicting quality trends under varying environmental conditions.

Step 5. Calculate and output evaluation metrics. In data analysis and modeling, particularly in fields such as regression analysis and chemometrics, the coefficient of determination R² (Formula (4)), root mean square error of calibration (RMSEC) (Formula (5)), and residual predictive deviation (RPD) (Formula (6)) are commonly used evaluation metrics.

$$R^2=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n}{(y_i-{\widehat{y}}_i)}^2}{{\displaystyle \sum _{i=1}^n}{(y_i-\overline{y})}^2}}$$

(4)

$$RMSEC=\sqrt{{\displaystyle \frac{{\displaystyle \sum _{i=1}^n}{(y_i-{\widehat{y}}_i)}^2}{n}}}$$

(5)

$$RPD={\displaystyle \frac{s}{\mathrm{RMSECV}}}$$

(6)

where n is the sample size, $y_i$ is the actual value of the i-th sample, ${\widehat{y}}_i$ is the predicted value of the i-th sample, $\overline{y}$ is the mean of all valid values, and s is the standard deviation of the samples.

Step 6. Output the prediction results and generate comparison plots of actual values versus predicted values. The value of R² ranges between 0 and 1. A value closer to 1 indicates a higher goodness-of-fit of the model to the data. RMSEC measures the average deviation between predicted values and actual values on the training set; a smaller value signifies higher prediction accuracy on the training set. RPD evaluates the predictive capability of the model. An RPD greater than 3.0 indicates strong predictive performance, with larger RPD values reflecting a stronger predictive capability.

(4)

Flexible wireless vibration sensing performance

Through simulating real-world table grape cold chain logistics environments, functional testing, data acquisition testing, and communication testing were conducted on the system. This comprehensive evaluation assessed the performance of the flexible wireless vibration sensor in data collection, processing, transmission, and system stability, while comparing it with traditional monitoring system to highlight its advantages in monitoring. Additionally, system usability was evaluated.

3. Results and Discussion

3.1. Cold-Chain Quality Data Analysis

Data analysis of the cold chain process for table grapes can elucidate quality changes, help optimize cold chain parameters and monitor quality, thereby providing a basis for cold chain monitoring procedures.

The acceleration in the X- and Y-directions exhibited significant variation trends. In contrast, changes in the Z-direction acceleration were relatively less pronounced, as shown in Figure 5a. This observation aligns with the force characteristics experienced by table grapes during actual cold chain transportation [28]. Figure 5b displays the sensor data stored on the Alibaba Cloud platform.

Acceleration changes across different directions varied under distinct vibration levels applied to table grapes. The experiment included six vibration levels, illustrated in Figure 5c–h. During transportation on flat roads (Levels 1–2), vibration amplitudes were minor: approximately 0.4 g for the X-axis, 0.125 g for the Y-axis, and a resultant vector acceleration of ~1.25 g. At the highest vibration level (Level 6), X-axis acceleration reached ~0.9 g, Y-axis reached ~0.25 g, and the resultant vector acceleration reached ~1.8 g.

Quality evaluation parameters for table grapes varied with vibration levels. In Figure 6a, the berry drop rate progressively increased with higher vibration levels, with its minimum, mean, median, and maximum values all showing upward trends. This indicates that elevated vibration levels exacerbate berry shedding. Conversely, flesh firmness (Figure 6b) and skin firmness (Figure 6c) decreased as the vibration intensity increased. This decline results from damage to the cellular structure during vibration, causing leakage of cellular contents and a reduction in firmness. The pedicle pull force (Figure 6d) also decreased at higher vibration levels. Continuous vibration induces mechanical stress at the pedicel-berry junction and within pedicel tissues, damaging vascular bundles and weakening intercellular adhesion.

The heatmap of chemometric parameters during the cold chain process (Figure 6e) reveals trends across vibration levels. For instance, gradually lighter colors corresponding to flesh firmness indicate declining values, reflecting cellular damage that may compromise textural properties.

3.2. Correlation Analysis Between Vibration Data and Cold-Chain Quality Damage

An in-depth analysis of the correlation between vibration data and quality deterioration is of great significance for optimizing grape cold chain logistics strategies, reducing losses, and enhancing economic benefits. Theoretical analysis can provide valuable references for research in this field.

Figure 7 presents the correlation validation results between vibration data and table grape cold chain quality indicators. Figure 7a shows that as vibration acceleration increases, the berry drop rate exhibits an overall upward trend, indicating that higher vibration acceleration exacerbates berry shedding. The Pearson correlation coefficient of 0.9288 confirms a significantly positive correlation between these two datasets. Conversely, Figure 7b–d demonstrate that with increasing vibration acceleration, the associated quality indicators (flesh firmness, skin firmness, pedicle pull force) display overall declining trends. This signifies that greater vibration acceleration intensifies negative impacts on these parameters. The Pearson correlation coefficients for these three figures are −0.9438, −0.9286, and −0.9449, respectively, confirming significantly negative correlations. The results indicate statistically significant correlations between vibration data and table grape cold chain quality indicators, supporting the feasibility of subsequent work on cold chain quality prediction.

3.3. Cold-Chain Quality Monitoring Model Evaluation

Establishing a quality monitoring model for the cold chain of table grapes is a crucial measure to ensure quality stability and reduce losses during the cold chain logistics process.

Figure 8a displays the original vibration data under different vibration levels, while Figure 8b presents the vibration data preprocessed using Z-score standardization. After preprocessing, the trends of the four vibration curves became more consistent compared to the raw data, with reduced value discrepancies, particularly at Levels 4–6.

Figure 8c–f presents the prediction results of the table grape quality during cold chain based on the MLR model. The results are shown in

Table 1. Prediction results of cold-chain quality parameters for fresh table grapes.

Parameter	Sample Size	Preprocessing Method	R2	RMSEC	RPD
Berry-drop rate	240	Z-score standardization	0.8997	0.0047	3.1917
Pulp firmness			0.8767	68.4084	2.7692
Peel hardness			0.8894	25.2233	2.8963
Peduncle tensile force			0.9155	0.4612	3.4764

. R² measures the goodness of fit of the regression model to the observed data, with values ranging from 0 to 1. A value closer to 1 indicates a better fit. All R² values exceed 0.87, indicating that the model effectively captures the relationship between the independent and dependent variables. RMSEC (Root Mean Square Error of Calibration) is the root mean square error of the training set, which measures the average deviation between the predicted and actual observed values. A smaller RMSEC value indicates higher prediction accuracy. The RMSEC values for shattering rate and pedicel tensile strength are relatively small, demonstrating good prediction performance. In contrast, the RMSEC values for flesh firmness and skin firmness are larger, which may be due to the unsuitability of the data type. RMSEC is more sensitive to larger errors, highlighting extreme prediction inaccuracies. RPD (Residual Predictive Deviation) reflects the relative relationship between the dispersion of model predictions and the prediction errors. A higher RPD value indicates a stronger ability of the model to distinguish between the true values of different samples. The RPD values for shattering rate and pedicel tensile strength both exceed 3, indicating strong predictive capability of the model. The RPD values for flesh firmness and skin firmness are 2.76 and 2.89, respectively, suggesting relatively satisfactory predictive performance.

Figure 8g,h illustrates the deep learning regression method and training process based on the Backpropagation Neural Network (BPNN) model. After standardizing the data, the dataset was divided into 70% for training and 30% for testing. A BPNN with 10 neurons was constructed, and a maximum of 1000 training epochs was set. Finally, the graphs and evaluation results were generated. During the training process, although the number of training epochs continuously increased, all three relevant curves showed a significant downward trend. The R² value was negative, indicating that the BPNN model performed poorly in this training session and requires careful examination of the data.

Table 2. Comparison of MLR, PLSR, and BPNN models for predicting fresh table grape cold-chain quality.

Parameter	Sample Size	Preprocessing Method	MLR	PLSR	BPNN
Berry-drop rate	240	Z-score Standardization	0.8997	0.8069	−1.0909
Pulp firmness			0.8767	0.7541	−0.4698
Peel hardness			0.8894	0.8219	−2.2659
Peduncle tensile force			0.9155	0.7683	−0.3131

compares prediction results for table grape cold chain quality using MLR, PLSR, and BPNN models, with R² as the evaluation metric. The R² values of the MLR model are generally approximately 12% higher than those of the PLSR model, whereas the BPNN model yields negative R² values. This demonstrates the superior predictive performance of the MLR model. This is because the PLSR model, when there is strong correlation among independent variables, extracts principal components to reduce dimensionality while considering the covariance between independent and dependent variables, making it suitable for handling multicollinearity issues. The BPNN model, through nonlinear transformations between multiple layers of neurons, can automatically learn complex patterns and relationships in the data, thus being applicable when the relationship between independent and dependent variables is intricate and the dataset is large. In summary, the MLR model demonstrates the highest prediction accuracy.

3.4. Flexible Wireless Vibration Sensing Performance Evaluation

The overall performance evaluation of the system is shown in

Table 3. System performance evaluation.

Evaluation items	Evaluation Metric	Comparison
		Traditional Monitoring System	Flexible Wireless Vibration Sensing
Data acquisition	Acceleration data acquisition accuracy	±(0.1% ~ 5%)	±2 g: ~0.0005 g/LSB ±4 g: ~0.001 g/LSB
	Temperature & humidity data acquisition accuracy	Temperature: ±(1~3) °C Relative humidity: ±(3~10)%	Temperature: ±0.2 °C Relative humidity: ±2%
Data processing	Data processing speed	Data points per second: tens to thousands (1 byte each)	Up to 150 Mbps
	Data processing accuracy	90% ~ 99%	>99%
Communication	Signal loss rate	1% ~ 10%	1%
	Communication range	10 m~5 km	70 m (open field)
System stability	Long-term operation stability	Depends on system type, application scenario, Technology level, and maintenance	Stable
	Stability under varying working conditions		Adapts well to diverse operating conditions
System usability	User interface friendliness	Subject to multiple factors, including design philosophy and technical level	Cloud dashboard clear; full-featured
	Ease of the system configuration		Code updates are easy; the platform functions complete

, The flexible wireless vibration sensor demonstrates outstanding performance in data acquisition and processing. Its acceleration data collection achieves higher precision across different ranges (approximately 0.0005 g/LSB at ±2 g and 0.001 g/LSB at ±4 g), with narrower and more defined error margins. This enables the system to capture vibration conditions during table grape cold chain transportation precisely. The high processing speed (up to 150 Mbps) enables the handling of large volumes of data within shorter timeframes, meeting the requirements for real-time monitoring and rapid response. Communication capabilities remain stable, maintaining reliable connectivity in open areas up to 70 m.

The system adapts well to diverse working conditions, demonstrating strong environmental resilience. It operates stably within complex and variable cold chain logistics environments, ensuring continuous and accurate monitoring data. Comprehensively, the flexible wireless vibration sensor holds significant application value for monitoring the quality of the table grape cold chain and exhibits potential for further optimization and enhancement.

4. Conclusions

This paper investigates the correlation between vibration data and quality deterioration of table grapes during cold chain logistics. Focusing on table grapes in the cold chain process, a flexible wireless vibration sensor was designed for monitoring their quality. Through the design and integration of hardware and software systems, as well as vibration data processing, the flexible wireless vibration sensor was developed and evaluated. A correlation model and an MLR (Multiple Linear Regression) model for quality monitoring during the cold chain process of table grapes were established. After Z-Score normalization, the model achieved R² values greater than 0.87 and RPD values greater than 2.7. Compared with other models, it demonstrated the best goodness of fit for monitoring the quality of table grapes in the cold chain. This study simulated the actual cold chain logistics environment of table grapes and conducted functional tests, data acquisition tests, and communication tests on the flexible wireless vibration sensor. Comprehensive tests showed that the sensor has significant advantages in monitoring the quality of table grapes in the cold chain, enabling real-time and accurate capture of quality change information. The results indicate that the flexible wireless vibration sensor exhibits strong stability and robustness in practical logistics applications, providing an efficient and intelligent solution for preventing losses in the cold chain transportation of fruits and vegetables.

Nevertheless, the system still has certain limitations in terms of optimizing sensing performance, expanding the scope of model applications, and conducting multi-factor comprehensive studies. Future research should be directed towards expanding sample diversity, integrating multi-source data, and increasing sample size for more in-depth investigations to further enhance the system’s performance and practicality. Additionally, particular attention should be paid to the impact of vibration-induced damage on grapes.