Biomimetic Octopus Suction Cup with Attachment Force Self-Sensing Capability for Cardiac Adhesion

Abstract

This study develops a biomimetic soft octopus suction device with integrated self-sensing capabilities designed to enhance the precision and safety of cardiac surgeries. The device draws inspiration from the octopus’s exceptional ability to adhere to various surfaces and its sophisticated proprioceptive system, allowing for real-time adjustment of adhesive force. The research integrates thin-film pressure sensors into the soft suction cup design, emulating the tactile capabilities of an octopus’s sucker to convey information about the contact environment in real time. Signals from sensors within soft materials exhibiting complex strain characteristics are processed and interpreted using the grey wolf optimizer-back propagation (GWO-BP) algorithm. The tissue stabilizer is endowed with the self-sensing capabilities of biomimetic octopus suckers, and real-time feedback on the adhesion state is provided. The embedding location of the thin-film pressure sensors is determined through foundational experiments with flexible substrates, standard spherical tests, and biological tissue trials. The newly fabricated suction cups undergo compression pull-off tests to collect data. The GWO-BP algorithm model accurately identifies and predicts the suction cup’s adhesion force in real time, with an error rate below 0.97% and a mean prediction time of 0.0027 s. Integrating this technology offers a novel approach to intelligent monitoring and attachment assurance during cardiac surgeries. Hence, the probability of potential cardiac tissue damage is reduced, with future applications for integrating intelligent biomimetic adhesive soft robotics.

Introduction

In contemporary surgery, the precise and stable fixation of surgical site coupled with real-time monitoring of patient status is extremely important. Current tissue stabilizers primarily use a negative pressure attachment system comprising plastic suction cups, hollow U-shaped arms, and pressure controllers.^1–3 These devices maintain a negative pressure of 300–400 mmHg during surgery to secure cardiac tissues.⁴ However, prolonged high negative pressure may induce hemodynamic instability, local myocardial ischemia, and tissue injury and elevate the risk of postoperative complications.^5–7 Therefore, developing a tissue stabilizer that can maintain stable fixation while reducing negative pressure and monitoring attachment status in real time is critical for reducing cardiac injury and the risk of postoperative complications.

Biomimetics is receiving increasing attention in the development of medical devices, particularly when designing tissue stabilizers, where inspiration from the natural world is leveraged to enhance device functionality and adaptability.^8,9 The suction cup of the octopus has emerged as a subject of considerable interest among researchers owing to its simple structure, adaptability, and exceptional proprioceptive capabilities. Octopuses can stick firmly to surfaces in complex underwater environments and subtly control the adhesive force of their suckers to avoid harming themselves or their prey during predation or movement.¹⁰ This capability arises from the meticulous coordination between the structure of the octopus’s sucker and its nervous system, which can detect changes in internal pressure and adjust accordingly.^11,12 Researchers have developed various sucker patches,^13–17 bionic graspers,^18–20 and high-adhesion underwater suction cups.^21,22 However, most studies concentrate on improving the suction force. Extensive research on the self-sensing adhesive force of suckers and their application in cardiac surgery stabilizers is still emerging, presenting a technological challenge and an opportunity for innovation.

Recent research advances include the development of sensory suckers combined with various sensors, such as those by Sareh et al.,²³ which use fiber Bragg grating sensors in artificial suckers. Sensory suckers developed by Huh et al.²⁴ integrate pressure sensors within suckers to detect contact. Similarly, Lee et al.²⁵ applied a conductive layer of carbon nanotubes to sense and detect objects’ relative position and balance. Moreover, Doi et al.²⁶ developed capacitive proximity sensors enabling automatic detection of the distance from the baseplate to the sucker surface. The investigations mentioned previously primarily focus on the role of suckers as graspers for contact recognition, with a greater emphasis on measuring lifting force and imbalance for assessing load during operational procedures, which is not suitable for stabilizing the heart. In contrast, this study uniquely concentrates on the real-time measurement of adhesive force, specifically targeting the immediate interactive forces between the sucker and the heart. This focus is critical for ensuring the sucker’s reliable attachment during surgery, directly impacting its ability to securely grasp cardiac tissue without causing damage.

On the contrary, in the nonlinear nature of the strain in soft materials, artificial neural networks serve as effective tools for interpreting these complex signals owing to their exceptional performance in signal processing and pattern recognition.^27–31 Therefore, combining sensor technology with artificial neural networks can mimic the self-sensing ability of natural organisms.

Building on the work of our research group,⁸ a flexible film sensor is integrated into the device to simulate the tactile sensory system of the octopus. This integration enables monitoring the contact pressure between the suction cup and the surface of the heart. Furthermore, a network model based on grey wolf optimization-back propagation algorithms (GWO-BP) was incorporated to analyze pressure signals from sensors embedded in soft materials. This biomimetic model combines flexible membrane sensors with the GWO-BP algorithm, replicates the self-sensing capabilities of octopus suckers, provides real-time feedback on adhesive force, and guides surgeons in precision operations. Figure 1 shows the design inspiration and general structure of the suction cup. This unique biomimetic approach is expected to significantly mitigate damage to cardiac tissue during surgery and enhance surgical outcomes. Although research in this field is still emerging, tissue stabilizers are expected to become increasingly intelligent with further technological advancement and in-depth study, serving the global medical health industry.

FIG. 1.

The overall design of an octopus-like suction cup with self-sensing function of soft tissue adsorption force. The image shows a new type of cardiac stabilizer that integrates a soft suction cup, inspired by the structure of the octopus suction cup. This device uses flexible electronic films to mimic the tactile feedback of the suction cup and incorporates neural networks to simulate the neural processing of sensory data, similar to how an octopus brain interprets information from its suction cup sensors. By leveraging this technology, the cardiac stabilizer can dynamically adjust its attachment force for optimal stability and support.

Materials and Methods

Flexible film sensors

Sensors are indispensable for gathering data from the environment and have numerous applications in fields such as medical health, human–computer interaction, and electronic skin. Applications such as octopus suction cups comprise organic silicone, hence flexible sensors³² are the optimal choice. They combine flexibility with sensitivity and adaptability while preserving the cup’s structural integrity and adherent function.

The DF9-40 series flexible thin-film pressure sensor from Suzhou Nengsihui Electronic Technology Co., Ltd. is chosen in this article. The sensor is prized for its remarkable thinness at only 0.05 mm. This feature, outlined in Figure 2d, and its softness ensure seamless and nonintrusive integration into the suction cup, leaving its natural performance unaffected. The sensor’s design, detailed in Figure 2a, comprises several layers: an adhesive for secure placement, a PVC plastic film for protection, a resistive carbon element acting as a pressure-sensitive component owing to its variable resistance, a spacer for structural support, and a silver conductor for signal transmission. This sensor operates on a resistance-based principle, where pressure application causes a significant and measurable change in the carbon ink’s electrical resistance, that is, the resistance decreases with increasing pressure. According to the circuit diagram in Figure 2b, an effective open circuit higher than 10 MΩ in a no-pressure state closes under pressure.

FIG. 2.

(a) Schematic diagram of thin-film pressure sensor structure. (b) Measuring circuit. (c) Flexible thin-film sensor size, unit: mm. (d) High-speed analog–digital (AD) microcontroller.

In our system, the pressure sensor is interfaced with a multichannel pressure conversion module (MY2901 model from Suzhou Nengsihui). This module translates the flexible thin-film pressure sensor’s analog resistance data into a digital format. The conversion to digital facilitates easier integration with neural network algorithms, allowing for more efficient training and learning processes within network models. This digital signal represents the analog-to-digital (AD) data. These data are transmitted to a computer via a serial port (Fig. 2c), allowing for efficient recording and analysis. Accurate and real-time pressure monitoring directly correlates with the AD voltage signal and the exerted pressure.

Preparation of octopus suckers

The proposed structural model integrates a design based on the imitation of octopus suction cup structures to replicate the function of octopus suction cups more closely, featuring unique hip protrusions (as highlighted in Wang’s article [8]). Therefore, it was ensured that the placement of flexible electronic sensors had a minimal impact on the integrity of the suction cup’s structure and its adhesion capability to simulate the sensory functions of real octopus suction cups. Three key locations were selected for sensor implantation, and the best implantation sites were selected based on the sensors’ responses. The sensor placement in this article aimed to ensure optimal performance, reflecting the cup’s interaction with its environment.

The key locations for sensor implantation are as follows: the top of the cup (near the pressure application point), the acetabulum region (playing a critical role in the cup’s ability to contract and generate suction), and the funnel region (closest position to the adsorbed object). These correspond to the upper, middle, and lower positions for sensor implantation as illustrated in Figure 3c.

FIG. 3.

The manufacturing process of self-sensing octopus suckers. (a) Pour the Smoothon silicone A and B into the container at a ratio of 1:1, stir thoroughly, and remove bubbles. (b) The bionic octopus sucker with embedded sensor is carried out by batch pouring. The sensor in the infundibulum is placed first and the mixed silica gel is poured into the infundibulum mold. After it is formed, the acetabulum and the top sensor are placed through the micro-holes, and the rest is injected with silica gel. (c) Tiny holes in the top and acetabulum of the sensor (SOLIDWORKS software mold and physical demonstration). (d) Drawings of the actual manufacturing process. (e) Sucker size.

Micro-holes were precisely engineered into the mold of the suction cup to ease the integration of the sensors without compromising the cup’s design. These holes are detailed in Figure 3d and were deliberately kept as small as possible to maintain the cup’s integrity. The micro-hole length shall not exceed 6 mm, and the width shall not exceed 0.3 mm.

The proposed suction cup is made of Smoothon silicone,³³ a common medical material with no irritation, toxicity, allergic reaction to human tissue, and very little rejection reaction. It is a common medical material. The sensor implantation process (Fig. 3) involves mixing Smoothon silicone components A and B in equal parts. Then, thorough stirring and degassing are conducted to remove air. The mixture is allowed to solidify after the sensors are placed in the lower mold’s infundibulum region and poured into the silicone. Subsequently, hip joint and apex region sensors are positioned into the upper mold’s micro-holes, and another silicone injection is applied. The suction cup is demolded after solidification, showcasing the finished product in Figure 3d. It should be noted that holes are reserved in the top of the suction cup. However, the top of the current using the suction cup is sealed. Moreover, the adsorption and detachment of the suction cup are completed by mechanical compression and pull-up to verify the basic function of the suction cup without a complete vacuum system.

Experiment

This section outlines the experimental framework, structured in three phases to support the research objectives. The initial phase involved conducting experiments on silicone-based substrates. These substrates’ varying thickness and flexibility allow for assessing sensor responses under different conditions. Following this, insights from the initial experiments guided the integration of sensors into suction cups, aiming to optimize sensor placement for accurate force monitoring during application. Finally, the GWO-BP algorithm was applied to predict the forces exerted, enhancing the design of the suction cup. This approach facilitated the translation of sensor data into practical insights, improving the precision and safety of the suction cup in real-world applications. All experiments were repeated 10 times and averaged.

Substrate model analysis and experiment

The sensor was placed at three locations of four different thicknesses of the silicon substrate to explore the effect of the sensor’s embedment depth on its performance, providing initial insights for us to understand and ultimately optimize the application of flexible thin-film sensors in more complex structures.

Substrate model analysis

As shown in Figure 4a, the flexible substrate is compressed through a tensile machine. When the sensor is located above the flexible substrate, the pressure on the sensor ( $P_{u}$ ) can be expressed as follows: $P_{u} = P_{f} - P_{e}^{u}$ (1)where $P_{f}$ is the tensile machine’s force and $P_{e}^{u}$ denotes the component of the applied force that is counterbalanced by the internal elastic potential energy within the substrate.

FIG. 4.

The sensor-embedded substrate simulates the suction cup compression experiment. (a) Schematic diagram of the substrate model for simulating the chuck built-in sensor. (b) Use the tensile machine to press the substrate, and collect the data experiment diagram through the conversion module and the host computer. (c)–(e) Experimental data diagrams of substrates fabricated using 00-30, 00-50, and 10A materials, respectively.

When the sensor is located beneath the flexible substrate, the pressure on the sensor ( $P_{l}$ ) can be expressed as follows: $P_{l} = P_{f} + P_{s} - P_{e}^{l}$ (2)where $P_{s}$ is the substrate’s gravitational force on the sensor and $P_{e}^{l}$ denotes the component of the applied force that is counterbalanced by the internal elastic potential energy within the substrate.

When the sensor is located in the middle of the flexible substrate, the pressure on the sensor ( $P_{c}$ ) can be expressed as follows: $P_{c} = P_{f} + {\frac{1}{2} P}_{s} - P_{e}^{c}$ (3)where $P_{e}^{c}$ denotes the component of the applied force that is counterbalanced by the internal elastic potential energy within the substrate.

To indicate variations in the effect of the internal elastic potential energy caused by the embedded position of the sensor, distinct superscripts are utilized. $P_{s}$ can be disregarded owing to the restricted volume of the square flexible substrate and the small area of the flexible film sensor (≈0). For the sensor placed at the lower and upper positions of the flexible substrate, owing to the reciprocity of forces, $P_{e}^{l} = P_{e}^{u}$ , as per formulas (1) and (2), which leads to $P_{u} \approx P_{l}$ .

Conversely, the elastic modulus experiences an increase in value when the sensor is centered on the flexible substrate as opposed to the upper and lower extremities; this leads to $P_{e}^{c} > P_{e}^{l} = P_{e}^{u}$ . Consequently, we can deduce that: $P_{u} \geq P_{l} > P_{c}$ (4)

This study aimed to determine the optimal placement of sensors on a flexible substrate and achieve the highest sensitivity—a crucial aspect for various applications, including biomedical devices. The theoretical analysis predicted that the sensor in the upper and lower positions performs better than the middle position, which will be experimentally validated.

Substrate model experiment

Three types of silicone materials were used in the study—Ecoflex 00–30, Ecoflex 00–50, and Dragon Skin 10—to fabricate flexible substrates of varying thicknesses (2 mm, 4 mm, 6 mm, and 8 mm), positioned at different locations (upper, middle, and lower) for sensor testing. Various thicknesses were chosen to evaluate the impact of material thickness on sensor placement and its performance, including sensitivity, material elasticity, and signal transmission efficiency. Moreover, various thicknesses were also chosen to simulate the operating conditions similar to those of an octopus sucker’s nervous system. The experiments were conducted using a uniaxial tensometer provided by Tohnichi Instruments Ltd. and equipped with a high-precision TRANSCELL tension sensor. A custom-designed compression tool (Fig. 4b) was used with the tensometer to apply controlled pressure to the embedded sensors within the flexible substrates. Hence, the influence of variations in substrate thickness on the performance of the sensing system was investigated.

The experimental results are shown in Figure 4c–e, which are the experimental data of the sensor placed on the upper part (black line), the middle part (red line), and the lower part (green line) of the flexible substrate. The sensor was placed on the rigid bench for direct pressure test to explore the influence of flexible materials on the sensor’s performance. The results are represented by the blue line. It can be seen from Figure 4 that under the same thickness, the sensor response increases with the increase of material hardness. Under the same material, the sensor response decreases with the increase of the thickness of the flexible substrate; especially the sensor in the middle position is the most affected. Therefore, the sensor embedment depth is small, and the performance is better.

Experiment with a built-in sensor in an octopus sucker

In this section, we embed the sensor in three locations of the sucker, the top, acetabulum, and infundibulum, to determine the location of the sensor embedment. The experiments were conducted under two conditions: on a standard spherical surface made of a resinous material and on a biological (pig heart) surface.

Standard spherical surface

In this study, the optimal location for embedding sensors within a suction cup was identified by examining the response at three specific sites: the top, the acetabulum (middle), and the infundibulum (lower region). Tests were conducted to determine whether the hardness of various materials, specifically Ecoflex 00–30, Ecoflex 00–50, and Dragon Skin 10A, influences the selection of the sensor location.

The conducted experiments used a single-axis tensile machine provided by Dongri Instrument Co., Ltd., equipped with a highly accurate TRANSCELL tensile pressure sensor and a LINE photoelectric encoder to measure displacement with high precision. This setup allowed us to apply controlled pressure, as shown in Figures 5a and 5b, and monitor the corresponding sensor response via a computer interface capturing AD signal values. To bolster the reliability of our findings, each experiment was replicated 10 times.

FIG. 5.

Experimental diagram of suction cup. (a) Drawing machine compression suction cup to obtain the overall data experimental picture. (b) Sequential photo showing the variation of the stretching machine continuously pressing down on the suction cup. (c) Experimental result diagram and box pattern diagram of three special pressure points (45 N, 50 N, and 55 N) in 10 experiments.

Figure 5c displays the experimental results, showing that the sensor’s best performance is consistently at the lower position within the anticipated working range (45–55 N), followed by the upper position, with the middle position performing the worst to maintain the same working pressure as a cardiac stabilizer.

On the 00-30 material, we selected three representative pressure values (45 N, 50 N, and 55 N) and depicted the collected AD values from 10 experiments in a boxplot in Figure 5c, which more clearly demonstrates the stability of the sensor’s output.

Table 1 provides a detailed statistical analysis of the experimental data, including the sensor AD values’ mean deviation, standard deviation, and range. The analysis indicates that the mean deviation of the sensor AD values lies between 0.2 and 2.99, the standard deviation ranges from 0.25 to 3.87, and the range is 1.73–22.72. These statistical parameters offer quantitative support for the obtained conclusion that the sensor exhibits good performance within the predetermined working range.

Table 1.

Experimental Statistics on a Standard Sphere

	00-30			00-50			10A
	Lower	Upper	Center	Lower	Upper	Center	Lower	Upper	Center
AD	1.32	1.12	0.20	2.99	2.73	0.52	2.79	1.62	0.20
STD	1.67	1.45	0.25	3.87	3.48	0.68	3.65	2.12	0.27
R	8.21	9.97	1.73	17.93	15.41	4.32	22.72	15.15	2.45

AD, average deviation; R, range; STD, standard deviation.

Biological surface

This segment describes the implementation of soft octopus suckers that are equipped with embedded sensors and have Shore hardness values of 00-30, 00-50, and 10A. To monitor and quantify real-time pressure, these suction devices were adhered to the surface of a porcine heart. The objective was to identify the optimal location for sensor embedding that would be most practical for use in cardiac operations. The schematic representation of the experimental setup is shown in Figure 6.

FIG. 6.

Experiments on surface suckers of living organisms (pig hearts). (a) Overall diagram of experiment. (b) The results of 10 experiments. (c) Some special points of the box diagram.

The results of 10 experiments are shown in Figure 6b and consist of data obtained via sensors integrated into suction cups made of three different materials. Notably, the sensor positioned highest demonstrated the highest level of performance, with the lower sensor closely trailing behind. In contrast, the middle sensor exhibited a comparatively reduced sensitivity to changes in pressure. This finding exhibits a marginal deviation from prior research, the discrepancy may be attributed to the experimental setup, particularly the differences in the sensor’s contact surfaces. The upper sensor’s contact interface was in direct contact with a rigid probe. In contrast, the lower sensor interfaced with flexible materials, namely, silicone suction cups and cardiac tissue. Such variations in contact configuration could potentially introduce measurement errors. The limitations of these findings must be acknowledged. Hence, a vacuum pump will be used in future research to generate negative pressure and more accurately evaluate the suction cups’ performance. The current results should primarily serve as a reference to lay the groundwork for future research and refinement of experimental methods.

Figure 6c shows the box diagram drawn by AD values collected in 10 experiments for suckers made of 00-30 materials, which more intuitively shows the stability of sensor output. Table 2 compiles the statistical analysis of experiments conducted on porcine hearts. The summary indicates that the sensor’s average deviation lies between 2.64 and 6.06, the standard deviation ranges from 3.33 to 7.92, and the range is 10.55–26.04. Figure 6c presents a boxplot of 10 AD values at pressure levels of 45 N, 50 N, and 55 N on the 00-30 material, offering a more visual representation of the sensor output’s stability.

Table 2.

Experimental Statistics on a Biological (Pig Heart) Surface

	00-30			00-50			10A
	Lower	Upper	Center	Lower	Upper	Center	Lower	Upper	Center
AD	4.40	4.92	2.68	3.66	6.06	2.64	2.59	4.05	2.88
STD	5.88	6.45	3.43	4.78	7.92	3.38	3.33	5.10	3.79
R	19.78	20.50	10.87	15.69	26.04	10.62	10.55	15.69	12.53

AD, average deviation; R, range; STD, standard deviation.

Based on the experimental results from “Substrate Model Analysis and Experiment” and “Experiment with a Built-in Sensor in an Octopus Sucker” section, combined with the previous adsorption simulation analysis based on finite element in our research group, the sensor is embedded at a depth of 1 mm and works best when embedded at the top of the sucker.

Pressure and attachment force prediction model based on GWO-BP algorithm

Compression and pull-off tests were conducted on the suction cup with embedded sensors at the top, extracting key features from the sensor data that reflect the adhesion state. By using the GWO-BP algorithm model, the complex relationships within the data were automatically learned and predicted. The analysis results were then used to interpret the performance and state of the suction cup, aiding physicians in adjusting the stabilizing frame promptly to ensure stable and soft fixation of the heart during the operation.

Experimental data collection

The flexible membrane sensor is integrated on the top of the suction cup. Moreover, an experiment of push–pull with the suction cup is carried out (Figs. 7a and b). The experimental results are shown in Figure 7c. Each experimental condition was rigorously tested in 10 independent trials.

FIG. 7.

(a) Suction cup compression process. (b) Suction cup pull-up process. (c) Tension-compression experimental data of suction cups made of three materials (00-30, 00-50, 10A).

The experiment involves a stationary fixture that supports the suction cup, whereas surface contact is adjusted via a remote control using a screw. Upon achieving zero pull pressure and sensor values, the screw induces an internal–external pressure differential that produces adhesion force (see Fig. 7a). As the stretching machine descends, the air is expelled, resulting in the formation of a vacuum. As the suction cup ascends, the pressure difference deforms it gradually until it detaches, thereby disrupting the vacuum (see Fig. 7b).

The outcomes (Fig. 7c) depict suction cups constructed from the following three substances: 00-30, 00-50, and 10A. As the stretching machine decreases in pressure, there is an increase in the pressure of the suction cup. As illustrated in Figure 7a, the force F1 during compression is denoted by the black line. Concurrently, sensor data, represented by the red line containing AD values, are transmitted to the computer. These lines exhibit a distinct correlation, which serves as the foundation for forthcoming neural network-based predictions.

As the stretching device ascends, the pressure of the suction cup decreases gradually. The dashed line in Figure 7b represents the force F2 that occurs during the upward pull. This pressure initially decreases until equilibrium is reached at 0 N. Constant upward force interacts with the adhesive force of the suction cup. When the upward force is outdone by the adhesive force, the suction cup maintains its attachment while experiencing a minor deformation. F2 continues to decrease until it reaches the pull-out point, which signifies the adhesive force at its maximum. Once the pulling force surpasses the capacity of the suction cup, the vacuum ceases to exist and F2 begins to increase. Eventually, the suction cup detaches and the sensor reading corresponds to the weight of the suction cup.

Data processing

The thin-film pressure sensor is attached to the top position of the soft suction cup to continuously monitor the force on the suction cup, providing continuous pressure data during the experiment. To digitize the analog signals obtained from the flexible film sensors, the MY2901 multichannel pressure conversion module was used. The specific model that was used possesses eight channels and is capable of directly receiving the converted digital signal through the communication interface. We extracted the pertinent 2-bit AD data from the 8-bit data stream for the two sensors embedded in the suction cups during the data processing phase.

To mitigate interference from the experimental environment and enhance the quality of our sensor data, we implemented the B-spline data interpolation technique. This approach eliminates outliers and smooths the sensor data, thereby enhancing the precision and dependability of the acquired data. The comprehensive preprocessing of the data is crucial to guarantee the robustness and integrity of our experimental findings.

GWO-BP algorithm

This research article presents a novel methodology for time series forecasting that integrates the back propagation neural network (BPNN) with the GWO algorithm. The pseudocode is presented in Table 3. This methodology handles AD-labeled input values and two sets of output values, F1 and F2. The primary objective of this fusion is to enhance the precision and performance of BP neural networks by using the optimization capabilities of the GWO algorithm.

Table 3.

Pseudocode of GWO-BP Algorithm

Initialize the grey wolf population.
Initialize the neural network with random weights and biases.
Set the maximum number of iterations or convergence criteria.
Set learning rate and other hyperparameters for backpropagation.
Initialize the positions of grey wolves.
Initialize the best positions for each grey wolf.
Initialize the global best position.
For each grey wolf:
Calculate fitness by training the neural network with its position.
If fitness is better than the best fitness:
Update the best fitness and best position.
Set the global best position to the best position among all grey wolves.
While not converged or maximum iterations reached:
For each grey wolf:
Update the position based on GWO rules.
For each grey wolf:
Update the neural network weights and biases based on their position.
For each grey wolf:
Calculate fitness by training the neural network with its updated weights and biases.
If fitness is better than the best fitness:
Update the best fitness and best position.
Set the global best position to the best position among all grey wolves.
Return the global best position, which contains optimized weights and biases for the neural network.

GWO-BP, grey wolf optimizer-back propagation.

To effectively train the neural networks and optimize the performance of the GWO-BP algorithm, a well-documented training dataset is required. The experimental data used in this study were primarily obtained from the data collection process, described in the “Experimental Data Collection” section. The input of this dataset comprises AD values obtained from the sensor, whereas the output variables of interest are denoted as F1 and F2.

The GWO-BP algorithm commences in this pseudocode by initializing grey wolves, their positions, and a neural network endowed with arbitrary weights and biases. Subsequently, the GWO algorithm is used to iteratively modify the grey wolf positions. By these updated positions, the weights and biases of the neural network are adjusted. The fitness is computed through the training of the neural network. In accordance, the best positions and global best positions are updated.

These steps are repeated until convergence occurs or the algorithm reaches its maximum number of iterations. In conclusion, it provides the optimized weights and biases for the neural network as the global best position.

This study presents a novel GWO-BP hybrid algorithm specifically designed for time series forecasting. The algorithm uses AD values as inputs and generates dual sets of output forces, denoted as F1 and F2. Significant progress has been made in prediction accuracy, convergence optimization, and heightened generalization capabilities through the seamless integration of the GWO algorithm and the BPNN. The potential of this integration is considerable across various domains in which accurate time series prediction is critical.

Experiments and results

A training set of 400 instances was used to optimize the model’s parameters, whereas a validation set consisting of 72 instances was used to evaluate the model’s performance. We monitored the loss function during multiple iterations of the GWO-BP algorithm to ensure that the network weights and thresholds were being updated continuously. Mean absolute error (MAE), mean squared error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE) were used to assess the performance of the GWO-BP algorithm in predicting pressure. These metrics quantified the discrepancies between the predicted and the actual pressure data. Each of the 30 iterations of the experiment was accompanied by the time (T) required to predict a solitary data point.

Discussion and analysis

The performance of the GWO-BP algorithm was evaluated using the following five metrics: MAE, MSE, RMSE, MAPE, and the time (T) needed to predict a given set of data points (Table 4). Figure 8 shows the suction cup data made of 00-50 materials. The comparison and error between the predicted and the actual results (F1 and F2 pressure) are illustrated in Figures 8(a) and 8(b). Figures 8(c) and 8(d) are box plots of F1 and F2, respectively. Figure 8(e) depicts the convergence of the GWO-BP algorithm, whereas Figure 8(f) illustrates the duration required to generate 30 predictions.

FIG. 8.

Experimental result. (a) Forecast results and error statistics of F1 and F2. (b) The result box diagram of F1 and F2 is running 30 times. (c) Convergence diagram of GWO-BP algorithm. (d) The algorithm runs 30 times to predict the time required for a single set of points. GWO-BP, grey wolf optimizer-back propagation.

Table 4.

Experimental Results of F1 and F2

	F1				F2
	MAE	MSE	RMSE	MAPE(%)	MAE	MSE	RMSE	MAPE(%)	T(s)
Mean	0.2649	0.1266	0.3508	4.9264	0.0795	0.0109	0.1040	4.8361	0.0027
Max	0.3541	0.2028	0.4503	8.1354	0.1016	0.0165	0.1285	11.8634	0.0040
Min	0.2026	0.0593	0.2436	3.0074	0.0694	0.0077	0.0878	0.9715	0.0015
Median	0.2527	0.1189	0.3448	4.6942	0.0775	0.0106	0.1030	3.8072	0.0026
Variance	0.0014	0.0019	0.0037	1.5882E-4	5.0798E-5	3.8608E-6	8.5353E-5	8.0346E-4	4.37208E-7

MAE, mean absolute error; MAPE, mean absolute percentage error; MSE, mean squared error; RMSE, root mean square error; T, time.

MAE, MSE, and RMSE evaluate the precision and accuracy of predictions. An average prediction error of a smaller magnitude is indicated by a lower MAE, whereas error spread is denoted by MSE and RMSE; smaller values are preferable. The mean error for F1 is 0.2649 N and for F2 it is 0.0795 N, both of which fall within an acceptable margin of error, based on 30 experiments. The accuracy of predictions is quantified by MAPE as a percentage of the actual pressure values. A reduced MAPE signifies enhanced precision. The fact that the mean MAPE for F1 is 4.9264% and for F2 is 4.8361% demonstrates the quality of the model.

Considerable importance is placed on the computational time necessary for the GWO-BP algorithm to forecast individual datasets, particularly when time-sensitive or real-time applications are involved. The experimental scene described in this article places significant emphasis on real-time performance. The mean time required to predict F1 and F2 is 0.0027 s, which is sufficient to satisfy real-time demands.

Upon assessing MAE, MSE, RMSE, and MAPE across numerous experiments, we discovered that their performance was consistent and favorable, indicating that the GWO-BP algorithm is dependable for pressure prediction. Precise predictions are achieved owing to the effective capture of pressure patterns exhibited by metrics with low values. The iterative convergence of metrics serves as evidence that parameter optimization was successful. Because of its computational efficiency, the algorithm applies to real-time scenarios. In conclusion, the GWO-BP algorithm is computationally efficient and generates precise pressure forecasts with low MAE, MSE, RMSE, and MAPE values. Additional investigation might examine the performance of the method on more extensive and intricate datasets to evaluate its scalability and generalizability.

Conclusion

This study introduced a novel soft octopus suction device with self-sensing capabilities to enhance cardiac surgery’s precision and safety. The device’s real-time ability to monitor its attachment status offers surgeons a critical tool for minimizing the risk of heart tissue damage during procedures requiring secure tissue adhesion.

The research started with evaluating thin film sensors embedded in flexible materials based on their performance and variations in material, thickness, and positional placement. The results showed that the sensors located at the top and top of the flexible substrate are significantly better than those placed in the middle. Subsequently, in the standard spherical sphere experiment, the funnel sensor performed best, whereas in the biological tissue experiment, the top position sensor performed best. Combined with the results of the previous Finite Element Method (FEM) simulation analysis, the sensor is embedded at a depth of 1 mm and works best when embedded at the top of the sucker.

Using a soft suction cup with a sensor embedded at the top, the key features of sensor information related to the attachment state of the sucker were extracted via compression pull-off experiments. The adhesion of soft suction cups is automatically identified and predicted by the GWO-BP algorithm model. The algorithm’s error rate is lower than 0.9715%, and the average time for predicting adhesion is 0.0027 s, meeting the real-time operation requirements. This accuracy and real-time performance are essential for surgical heart surgery.

Future work will focus on improving the performance consistency of the sensors and exploring the detailed composition and optimization strategies of neural network models to mimic the neural processing capabilities of the octopus. This will include the simulation of a basic form of “local refinement” and decision-making processes aimed at real-time suction and grip strength adjustment under dynamic surgical conditions, enhancing applicability and precision during surgery. These efforts are expected to enhance the reliability of bionic soft suction cup technology in clinical applications, thereby advancing its practical application in medical surgery.

Footnotes

Authors’ Contributions

Z.W.: Review and editing (lead), writing—original draft (lead), formal analysis (lead), and software (lead). G.S.: Conceptualization (lead) and writing—review and editing (equal). X.F.: Software (supporting). P.X.: Methodology (supporting). L.Z.: Conceptualization (supporting) and writing—original draft (supporting).

Author Disclosure Statement

The authors declared that they have no conflict of interest and do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.

Funding Information

No funding was received for this article.

References

Medtronic showcases Octopus2 tool for beating heart surgery. Journal of Clinical Engineering, 1999; 24(5):296–299.

Razuk Filho

, Dos Santos

, Dos Reis

, et al. 217.8: How to do it: Use of octopus tissue stabilizer for minimal manipulation approach of the bronchial anastomosis in lung transplant. Transplantation, 2022; 106(9S):S65–S65.

Razuk

, Santos

SLD

, Reis

FPD

, et al. Use of octopus™ tissue stabilizer for minimal manipulation approach of bronchial anastomosis in lung transplant. Braz J Cardiovasc Surg, 2023; 38(6):e20220413.

Ammannaya

GKK

, Basantwani

, Mishra

, et al. Effect of octopus tissue stabilizer on cardiac output during off-pump coronary artery bypass graft surgery. Kardiochir Torakochirurgia Pol, 2019; 16(2):69–73.

Khatib

, Boettcher

, Freed

, et al. Acute, severe pulmonary arterial hypertension during off-pump coronary artery surgery: Is new myocardial ischemia, cardiac repositioning, or external mitral valve compression the culprit? Journal of Cardiothoracic and Vascular Anesthesia, 2016; 30(6):1744–1747.

Kinjo

, Tokumine

, Sugahara

, et al. Unexpected hemodynamic deterioration and mitral regurgitation due to a tissue stabilizer during left anterior descending coronary anastomosis in off-pump coronary artery bypass graft surgery. Ann Thorac Cardiovasc Surg, 2005; 11(5):324–328.

Pasarad

, Gopivallabha

, Singh

, et al. Left ventricular dissecting hematoma caused by tissue stabilizer. Sociedade Brasileira de Cirurgia Cardiovascular, 2020(3).

Wang

, Sun

, He

, et al. Octopus-inspired sucker to absorb soft tissues: Stiffness gradient and acetabular protuberance improve the adsorption effect. Bioinspir Biomim, 2022; 17(3):036005.

Kotak

, Maxson

, Weerakkody

, et al. Octopus-Inspired muscular hydrostats powered by twisted and coiled artificial muscles. Soft Robot, 2023.

10.

Graziadei

PPC

, Gagne

. Sensory innervation in the rim of the octopus sucker. J Morphol, 1976; 150(3):639–679.

11.

Albertin

, Simakov

, Mitros

, et al. The octopus genome and the evolution of cephalopod neural and morphological novelties. Nature, 2015; 524(7564):220–224.

12.

Birch

, Schnell

, Clayton

. Dimensions of animal consciousness. Trends Cogn Sci, 2020; 24(10):789–801.

13.

Baik

, Kim

, Park

, et al. A wet-tolerant adhesive patch inspired by protuberances in suction cups of octopi. Nature, 2017; 546(7658):396–400.

14.

Luo

, Klein Cerrejon

, Römer

, et al. Boosting systemic absorption of peptides with a bioinspired buccal-stretching patch. Sci Transl Med, 2023; 15(715):eabq1887.

15.

Ahmad

NFN

, Ghazali

NNN

, Rani

ATA

, et al. Bioinspired adhesive patch with octopus vulgaris micro-sucker and hexagonal tree-frog pad structures. Materials Science in Semiconductor Processing, 2023; 166:107731.

16.

, Hu

, Liu

, et al. 3D light-curing printing to construct versatile octopus-bionic patches. J Mater Chem B, 2023; 11(22):5010–5020.

17.

Lee

, Um

, Lee

, et al. Octopus‐inspired smart adhesive pads for transfer printing of semiconducting nanomembranes. Adv Mater, 2016; 28(34):7457–7465.

18.

Kennedy

EBL

, Buresch

, Boinapally

, et al. Octopus arms exhibit exceptional flexibility. Sci Rep, 2020; 10(1):20872.

19.

Xie

, Domel

, An

, et al. Octopus arm-inspired tapered soft actuators with suckers for improved grasping. Soft Robot, 2020; 7(5):639–648.

20.

Xie

, Yuan

, Liu

, et al. Octopus-inspired sensorized soft arm for environmental interaction. Sci Robot, 2023; 8(84):eadh7852.

21.

Frey

, Haque

ABMT

, Tutika

, et al. Octopus-inspired adhesive skins for intelligent and rapidly switchable underwater adhesion. Sci Adv, 2022; 8(28):eabq1905.

22.

Hwang

, Lee

, Kim

, et al. Soft microdenticles on artificial octopus sucker enable extraordinary adaptability and wet adhesion on diverse nonflat surfaces. Advanced Science, 2022; 9(31):2202978.

23.

Sareh

, Althoefer

, Li

, et al. Anchoring like octopus: Biologically inspired soft artificial sucker. J R Soc Interface, 2017; 14(135):20170395.

24.

Huh

, Sanders

, Danielczuk

, et al. A multi-chamber smart suction cup for adaptive gripping and haptic exploration. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2021: pp. 1786–1793.

25.

Lee

, Baik

, Hwang

, et al. An electronically perceptive bioinspired soft wet-adhesion actuator with carbon nanotube-based strain sensors. ACS Nano, 2021; 15(9):14137–14148.

26.

Doi

, Koga

, Seki

, et al. Novel proximity sensor for realizing tactile sense in suction cups. 2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020: 638–643.

27.

Loo

, Ding

, Baskaran

, et al. Robust multimodal indirect sensing for soft robots via neural network-aided filter-based estimation. Soft Robot, 2022; 9(3):591–612.

28.

Yang

, Ouyang

, Zhu

, et al. A biosensing system employing nonlinear dynamic analysis-assisted neural network for drug-induced cardiotoxicity assessment. Biosens Bioelectron, 2023; 222:114923.

29.

LeCun

, Bengio

, Hinton

. Deep learning. nature, 2015; 521(7553):436–444.

30.

Wang

, Zhao

, Wang

. A flexible iontronic capacitive sensing array for hand gesture recognition using deep convolutional neural networks. Soft Robot, 2023; 10(3):443–453.

31.

Shan

, Zeng

, Liu

, et al. Artificial tactile sensing system with photoelectric output for high accuracy haptic texture recognition and parallel information processing. Nano Lett, 2022; 22(17):7275–7283.

32.

PSC

, Antwerp

WPV

. Method of fabricating thin film sensors: US19940212961. 2024, 01(29): US5391250A.

33.

Siegenthaler

, Künkel

, Skupin

, et al. Ecoflex and Ecovio: Biodegradable, Performanceenabling Plastics Synthetic Biodegradable Polymers. Springer: Berlin; 2011, 245: pp. 91–136.