Prosthetic socket automatic adjusting system based on pressure sensing

Abstract

The invention discloses a prosthetic socket automatic adjusting system based on pressure sensing, and the system comprises a pressure collection module which is used for collecting contact pressure data in a prosthetic socket and constructing a pressure diagram sequence; the neural switch network modeling module is used for constructing a neural switch network consisting of a plurality of switching units; the structure and parameter joint optimization module is used for performing joint optimization on a switching path structure, a gating module and a mixing proportion parameter in the neural switch network; the state processing and intermediate control information generation module is used for forming intermediate control information; the adjusting instruction generating and driving module is used for forming adjusting input data from the intermediate control information and the memory vector, and further converting the adjusting input data into an execution control signal to drive a receiving cavity execution mechanism to complete structural deformation; and the feedback acquisition and circulation control module is used for acquiring and feeding back data and executing the next round of structure adjustment process. The dynamic fitting adjustment of the artificial limb receiving cavity is realized, and the wearing comfort and the real-time response capability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of prosthetic limb receiving cavity adjustment, and in particular to a prosthetic limb receiving cavity automatic adjustment system based on pressure sensing. Background Art

[0002] Currently, the prosthetic socket is a key structure that connects the human residual limb to the prosthetic device. Its fit directly affects the user's wearing comfort and gait stability. In traditional prosthetic systems, the size and structure of the socket are mostly customized by rehabilitation technicians based on static models, and it is difficult to automatically adjust according to real-time pressure changes during wearing. When the user is in different postures or activities, the contact pressure between the residual limb and the socket is prone to unevenness, causing tenderness, sliding, skin damage and other problems. Therefore, some studies have attempted to introduce pressure sensors to obtain pressure information inside the socket and fine-tune the structure through simple feedback logic. However, such systems generally lack the ability to deeply model the pressure distribution state, and the adjustment strategies are mostly based on rule control or heuristic settings, making it difficult to achieve accurate response in multi-time series and multi-state changes.

[0003] There are three common problems in the existing sensor feedback-based adjustment systems. First, pressure data are mostly processed using static analysis methods, lacking modeling of state evolution laws in time series, and unable to reflect the dynamic correlation of pressure changes during residual limb movement. Second, although a small number of studies have introduced neural networks for modeling, the network structure is mostly fixed, and the structural adaptability issues in task scenarios are not considered, resulting in overfitting or weak generalization performance of the model. Third, current structural adjustment decisions often rely on preset mappings or rule-triggered logic, lack the ability to adapt to different residual limb characteristics and behavioral changes, have coarse adjustment granularity and limited control accuracy, and cannot achieve the goal of fine and continuous automatic adjustment.

[0004] Therefore, how to provide an automatic adjustment system for a prosthetic socket based on pressure sensing is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] One purpose of the present invention is to propose an automatic adjustment system for a prosthetic receiving cavity based on pressure sensing. The present invention fully utilizes the dynamic timing modeling capability of the neural switch network and the structural search capability of the ice crystal ant colony optimization algorithm, and describes in detail how to model and adjust the pressure distribution in the prosthetic receiving cavity through mechanisms such as state gating, path exchange and parameter fusion. It has the advantages of continuous response, structural adaptability and high wearing comfort.

[0006] According to an embodiment of the present invention, a pressure-sensing-based automatic adjustment system for a prosthetic socket includes:

[0007] A pressure acquisition module is used to collect contact pressure data in the prosthetic socket and construct a pressure map sequence;

[0008] The neural switch network modeling module is used to construct a neural switch network composed of several switching units. Each switching unit is equipped with a spatial gradient gating module and a historical change gating module, and defines the mixing ratio parameters and the switching path structure.

[0009] a structure and parameter joint optimization module, for jointly optimizing the switching path structure, gating module, and mixing ratio parameters in the neural switch network using an ice crystal ant colony optimization algorithm to generate an optimized network structure configuration;

[0010] The state processing and intermediate control information generation module is used to perform state exchange based on the optimized network structure configuration, perform state fusion through the mixing ratio parameter, generate memory vectors, and form intermediate control information;

[0011] The adjustment instruction generation and driving module is used to convert the intermediate control information and the memory vector into adjustment input data, and further convert it into an execution control signal to drive the receiving cavity actuator to complete the structural deformation;

[0012] The feedback acquisition and loop control module is used to collect contact pressure data after structural deformation, form an updated pressure map sequence and input it into the neural exchanger network to execute the next round of structural adjustment process.

[0013] Optionally, modules can be connected using the following methods:

[0014] S1, collecting contact pressure data in the prosthetic socket and constructing a pressure map sequence;

[0015] S2. Constructing a neural switch network, wherein the neural switch network includes a plurality of switching units, each of which is provided with two gating modules, a spatial gradient gating module and a historical change gating module, for controlling whether the node state participates in the exchange, setting a mixing ratio parameter for fusing the exchanged state with the original state, and defining a switching path structure for transferring the activation state between time steps;

[0016] S3, using the ice crystal ant colony optimization algorithm to jointly optimize the switching path structure, gating module and mixing ratio parameters to obtain an optimized network structure configuration;

[0017] S4. Applying the optimized network structure configuration to the neural switch network to obtain an optimized network, inputting the pressure map sequence into the optimized network, performing state exchange between nodes that meet the gating conditions according to the exchange path structure, using the mixing ratio parameter to perform state fusion, and introducing the replaced state of the previous time step to form a memory vector to generate an intermediate control information sequence for control;

[0018] S5. Generate a receiving cavity adjustment instruction according to the intermediate control information sequence, and drive the actuator to generate structural deformation;

[0019] S6. Collect contact pressure data after structural deformation, construct an updated pressure map sequence, and input it into the optimization network to execute the next round of structural adjustment process.

[0020] Optionally, the spatial gradient gating module and the historical change gating module of each switching unit in the neural switch network of step S2 are used to jointly control whether the node state participates in the activation state exchange, specifically including:

[0021] Get the node i and its adjacent node set at the current time step t The pressure value p t (i) and p t (j), where Calculate the spatial gradient gate value G of node i s (i,t);

[0022] In the historical change gating module, the pressure value of node i in the past K time steps {p t-k (i)}, k=1,2,…,K, calculate the historical change gate value G h (i,t);

[0023] The spatial gradient gating value and the historical change gating value are the outputs of the gating function that changes with time. The spatial gradient gating parameter matrix and the historical change gating parameter matrix in the network are calculated. G s (i,t) and G h (i, t) and the set threshold value θ s and θ h Compare, if and only if G s (i,t)>θ s And G h (i,t)>θ h When both are true, the activation state of node i participates in the exchange operation.

[0024] Optionally, each switching unit in the neural switch network of step S2 sets a mixing ratio parameter for fusing the original activation state of the node at the current time step with the switching state passed in from the previous time step, specifically including:

[0025] At time step t, for node i that meets the exchange condition, let the original activation state be h t (i), the corresponding exchange state is the activation state h introduced from node j at time step t-1 t-1 (j), where And the gating condition is satisfied, that is, if and only if Gs (i,t)>θ s And G h (i,t)>θ h hour;

[0026] Assign a mixing ratio parameter α(i,t)∈[0,1] to node i to control the activation state fusion ratio;

[0027] Based on the mixing ratio parameters, calculate the fusion state of node i at time step t

[0028] Optionally, the switching path structure in the neural switch network of step S2 is used to specify the transfer mode of activation state between different time steps, specifically including: in time step t, constructing a three-dimensional structure path tensor W swap ∈{0,1}, where W swap [i,j,t]=1 means that node i receives the activation state from node j at time step t-1 at time step t. If W swap [i,j,t]=0 means that node i does not receive the activation state from node j at time step t-1 at time step t. For each node i that meets the gating condition, according to the adjacent node set With the structure path tensor, select the only exchange node Determine the exchange path weight ω(i,j,t)∈{0,1} so that Get the activation state h of node j at time step t-1 from the structure path tensor t-1 (j), and combined with time step t node h t (i), call the fusion state The fused state is output as the update of node i at the current time step.

[0029] Optionally, the S3 specifically includes:

[0030] S31, constructing an ant colony population set, where each ant individual represents a structural configuration, wherein the structural configuration includes a structural path tensor, a spatial gradient gating parameter matrix, a historical change gating parameter matrix, and a mixing ratio parameter;

[0031] S32, dividing the constructed pressure map sequence into a training set and a validation set, fitting the training set using the neural switch network under the current structural configuration, and performing the control information output process on the validation set, setting the fitness function as the difference score between the intermediate control information generated by the neural switch network on the validation set and the target control configuration, using the Manhattan distance function to compare the two, and using the score as an evaluation criterion for the quality of the structural configuration;

[0032] S33. In each round of iteration, the following optimization process is performed:

[0033] For the structure path tensor W swap The update introduces the structural inspiration factor η s (i, j, t) and the cross pheromone strength τ(i, j, t), by calculating the probability weight P(i, j, t) ∝ η s (i, j, t)·τ(i, j, t) determines the state exchange path configuration;

[0034] For the update of the spatial gradient gating parameter matrix, the historical change gating parameter matrix and the mixing ratio parameter α, a probabilistic perturbation strategy under the control of the lattice cooling mechanism is introduced: that is, according to the current temperature variable T k Control the perturbation amplitude of each parameter position Δ=∈·exp(-γT k ), where ∈ is the initial range of perturbation and γ is the cooling coefficient, which realizes the stable convergence of network structure and parameter space and compatibility with jump search;

[0035] S34. Record the structural configuration with the best fitness function value in all iteration rounds as the optimal structural configuration, which includes the optimized structural path tensor, spatial gradient gating parameter matrix, historical change gating parameter matrix and mixing ratio parameter, and apply the optimal structural configuration to the neural switch network.

[0036] Optionally, the S4 specifically includes:

[0037] S41, deploying the optimal structural configuration obtained by the ice crystal ant colony optimization algorithm to the neural switch network to form an optimized network with a fixed structure;

[0038] S42, inputting the constructed pressure map sequence into the optimization network, and the network sequentially extracting the node state as the initial activation state at each time step;

[0039] S43. In each time step, based on the state exchange path defined in the structural path tensor, the state relationship between the nodes is judged; when a node meets the activation conditions of its spatial gradient gate value and historical change gate value, and the value corresponding to the node in the current time step and the other node in the previous time step in the structural path tensor is 1, the current state is fused with the state of the associated node in the previous time step according to the mixing ratio parameter corresponding to the node;

[0040] S44. After the fusion process is completed, the replaced node state is saved as a historical version, the original state of the node in the previous time step is recorded as a memory vector, and the states of all nodes that have undergone state exchange and completed fusion in the current time step are combined with the corresponding memory vectors to form intermediate control information;

[0041] S45. Repeat the above state exchange and intermediate control information construction process until all time steps are traversed and a complete intermediate control information sequence is finally obtained.

[0042] Optionally, the S5 specifically includes: reading the intermediate control information output by the neural switch network, extracting the fusion state of each node and the corresponding memory vector, combining them in chronological order to form the input data required for adjustment, dividing the input data into control components corresponding to each receiving cavity adjustment unit, and converting them into data parameters indicating the action amplitude and execution timing of each adjustment unit, inputting the data parameters into the drive interface of the receiving cavity actuator to complete the structural deformation.

[0043] The beneficial effects of the present invention are:

[0044] First, by introducing the neural exchanger network as the core modeling structure and combining the spatial gradient gating module with the historical change gating module, it is possible to realize conditional dynamic state exchange between node states, and realize the fusion of the exchanged state and the original state through the mixing ratio parameter, which effectively improves the model's ability to represent the dynamic changes of multiple time steps in the receptor cavity pressure map sequence, and overcomes the shortcomings of the traditional network structure that is rigid and difficult to model the migration relationship of time series.

[0045] Secondly, the ice crystal ant colony optimization algorithm is used to jointly optimize the exchange path structure, gating parameters and mixing ratio parameters in the network. The path tensor is adjusted based on the structural inspiration factor and pheromone intensity, and the perturbation update of the gating and fusion parameters is controlled by the lattice cooling mechanism. This significantly enhances the structural adaptability of the network under different individual residual limb states and activity modes, realizes the task-adaptive configuration of the neural network structure, and avoids the problem of insufficient generalization ability caused by the fixed model structure.

[0046] In addition, the intermediate control information is used to drive the actuator of the socket to form structural deformation, and the feedback loop is combined to continuously collect and update the pressure map sequence, realizing a closed-loop dynamic adjustment process. This enables the prosthetic socket to respond and automatically adjust in real time according to the user's posture changes and usage scenarios, thereby improving the system's continuous adjustment capabilities and the user's wearing comfort experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 This is a flow chart of a method for an automatic adjustment system of a prosthetic socket based on pressure sensing proposed by the present invention;

[0049] Figure 2This is a schematic diagram of the module structure of a pressure-sensing-based automatic adjustment system for a prosthetic socket proposed by the present invention;

[0050] Figure 3 This is a joint optimization flow chart of the ice crystal ant colony optimization algorithm for the pressure-sensing-based prosthetic socket automatic adjustment system proposed in the present invention. DETAILED DESCRIPTION

[0051] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0052] refer to Figure 1-3 , a pressure-sensing-based automatic adjustment system for a prosthetic socket, comprising:

[0053] A pressure acquisition module is used to collect contact pressure data in the prosthetic socket and construct a pressure map sequence;

[0054] The neural switch network modeling module is used to construct a neural switch network composed of several switching units. Each switching unit is equipped with a spatial gradient gating module and a historical change gating module, and defines the mixing ratio parameters and the switching path structure.

[0055] a structure and parameter joint optimization module, for jointly optimizing the switching path structure, gating module, and mixing ratio parameters in the neural switch network using an ice crystal ant colony optimization algorithm to generate an optimized network structure configuration;

[0056] The state processing and intermediate control information generation module is used to perform state exchange based on the optimized network structure configuration, perform state fusion through the mixing ratio parameter, generate memory vectors, and form intermediate control information;

[0057] The adjustment instruction generation and driving module is used to convert the intermediate control information and the memory vector into adjustment input data, and further convert it into an execution control signal to drive the receiving cavity actuator to complete the structural deformation;

[0058] The feedback acquisition and loop control module is used to collect contact pressure data after structural deformation, form an updated pressure map sequence and input it into the neural exchanger network to execute the next round of structural adjustment process.

[0059] The present invention divides the system functions into six modules: pressure acquisition, modeling optimization, state processing, execution regulation and feedback control, clarifies the functional boundaries and maintains terminology consistency, and forms an overall control system architecture that can be implemented modularly. It has good engineering deployment and system expansion capabilities, which is conducive to industrial promotion.

[0060] In this embodiment, the modules are connected through the following methods:

[0061] S1, collecting contact pressure data in the prosthetic socket and constructing a pressure map sequence;

[0062] S2. Constructing a neural switch network, wherein the neural switch network includes a plurality of switching units, each of which is provided with two gating modules, a spatial gradient gating module and a historical change gating module, for controlling whether the node state participates in the exchange, setting a mixing ratio parameter for fusing the exchanged state with the original state, and defining a switching path structure for transferring the activation state between time steps;

[0063] S3, using the ice crystal ant colony optimization algorithm to jointly optimize the switching path structure, gating module and mixing ratio parameters to obtain an optimized network structure configuration;

[0064] S4. Applying the optimized network structure configuration to the neural switch network to obtain an optimized network, inputting the pressure map sequence into the optimized network, performing state exchange between nodes that meet the gating conditions according to the exchange path structure, using the mixing ratio parameter to perform state fusion, and introducing the replaced state of the previous time step to form a memory vector to generate an intermediate control information sequence for control;

[0065] S5. Generate a receiving cavity adjustment instruction according to the intermediate control information sequence, and drive the actuator to generate structural deformation;

[0066] S6. Collect contact pressure data after structural deformation, construct an updated pressure map sequence, and input it into the optimization network to execute the next round of structural adjustment process.

[0067] The present invention realizes dynamic modeling of the contact pressure diagram sequence of the prosthetic socket by constructing a neural switch network and introducing state gating and path exchange mechanisms, and forms a structural adjustment closed loop through feedback-driven actuators. It is significantly different from traditional prosthetic systems that rely on static rules and single-step perception. It has structural self-adaptation, timing perception and continuous adjustment capabilities, and can adapt to changing user states.

[0068] In this embodiment, the spatial gradient gating module and the historical change gating module of each switching unit in the neural switch network of step S2 are used to jointly control whether the node state participates in the activation state exchange, specifically including:

[0069] Get the node i and its adjacent node set at the current time step t The pressure value p t (i) and p t (j), where Calculate the spatial gradient gate value G of node i s(i,t), the expression is:

[0070]

[0071] This formula measures the "gradient strength" at a node in the spatial domain, that is, whether there is significant contact unevenness in the area where the node is located, and then determines whether to trigger a state exchange operation. Compared with the traditional global unified processing method, this method is position-sensitive and can significantly improve the model's recognition accuracy for areas with uneven pressure distribution, making state adjustments more targeted, avoiding unnecessary interference in stable areas, and improving regulation efficiency and system stability.

[0072] In the historical change gating module, the pressure value of node i in the past K time steps {p t-k (i)}, k=1,2,…,K, calculate the historical change gate value G h (i,t), the expression is:

[0073]

[0074] This formula enables the model to identify areas where pressure changes rapidly over time—specifically, locations where significant contact changes occur during residual limb movement. Compared to static threshold triggering, this approach provides time-awareness, enabling the adjustment strategy to respond to "changing trends." This helps the system intervene promptly when discomfort trends are detected, thereby improving system predictability and user comfort.

[0075] The spatial gradient gating value and the historical change gating value are the outputs of the gating function that changes with time. The spatial gradient gating parameter matrix and the historical change gating parameter matrix in the network are calculated. G s (i,t) and G h (i, t) and the set threshold value θ s and θ h Compare, if and only if G s (i,t)>θ s And G h (i,t)>θ h When both are true, the activation state of node i participates in the exchange operation.

[0076] The present invention introduces a spatial gradient gating module and a historical change gating module into each switching unit of the neural switch network, which can dynamically control whether the node participates in state exchange according to the spatial position relationship and historical change trend of the current node state, thereby improving the structural controllability and temporal consistency of the model and overcoming the problem of non-selective state diffusion in existing networks.

[0077] In this embodiment, each switching unit in the neural switch network of step S2 sets a mixing ratio parameter for fusing the original activation state of the node at the current time step with the exchange state passed from the previous time step, specifically including:

[0078] At time step t, for node i that meets the exchange condition, let the original activation state be h t (i), the corresponding exchange state is the activation state h introduced from node j at time step t-1 t-1 (j), where And the gating condition is satisfied, that is, if and only if G s (i,t)>θ s And G h (i,t)>θ h hour;

[0079] Assign a mixing ratio parameter α(i,t)∈[0,1] to node i to control the activation state fusion ratio;

[0080] Based on the mixing ratio parameters, calculate the fusion state of node i at time step t The calculation expression is:

[0081]

[0082] This formula introduces a dynamically adjustable mixing ratio parameter into its design, making the fusion process locally controllable and time-sensitive. Compared to traditional state-overlay exchange mechanisms, this fusion strategy not only preserves the node's own continuity memory but also fully incorporates state transition relationships across time steps. This improves the network's ability to express time-dependent features and its regulatory stability, avoiding information loss and mutations. It also provides smoother and more robust activation state input for system regulation, forming a key structural foundation for achieving dynamic state adaptation.

[0083] The present invention introduces a mixing ratio parameter into the state exchange mechanism, and fuses the exchange state and the original state with a proportional weight, thereby realizing a compatible processing method of smooth state transition and local disturbance regulation. Compared with a single replacement path exchange structure, it can effectively alleviate the risk of information loss and improve the stability and representation ability of the network in the time series modeling process.

[0084] In this embodiment, the switching path structure in the neural switch network of step S2 is used to specify the transmission mode of the activation state between different time steps, specifically including: in time step t, constructing a three-dimensional structure path tensor W swap ∈{0,1}, where W swap [i,j,t]=1 means that node i receives the activation state from node j at time step t-1 at time step t. If W swap[i,j,t]=0 means that node i does not receive the activation state from node j at time step t-1 at time step t. For each node i that meets the gating condition, according to the adjacent node set With the structure path tensor, select the only exchange node Determine the exchange path weight ω(i,j,t)∈{0,1} so that Get the activation state h of node j at time step t-1 from the structure path tensor t-1 (j), and combined with time step t node h t (i), call the fusion state The fused state is output as the update of node i at the current time step.

[0085] The present invention defines a clear exchange path structure and controls the state transfer relationship between different nodes through structural weights. Compared with the unstructured full connection mechanism in existing neural networks, this path structure provides structured support for the selective propagation of temporal activation states, thereby improving the efficiency and interpretability of the model.

[0086] In this embodiment, S3 specifically includes:

[0087] S31, constructing an ant colony population set, where each ant individual represents a structural configuration, wherein the structural configuration includes a structural path tensor, a spatial gradient gating parameter matrix, a historical change gating parameter matrix, and a mixing ratio parameter;

[0088] S32, dividing the constructed pressure map sequence into a training set and a validation set, fitting the training set using the neural switch network under the current structural configuration, and performing the control information output process on the validation set, setting the fitness function as the difference score between the intermediate control information generated by the neural switch network on the validation set and the target control configuration, using the Manhattan distance function to compare the two, and using the score as an evaluation criterion for the quality of the structural configuration;

[0089] S33. In each round of iteration, the following optimization process is performed:

[0090] For the structure path tensor W swap The update introduces the structural inspiration factor η s (i, j, t) and the cross pheromone strength τ(i, j, t), by calculating the probability weight P(i, j, t) ∝ η s (i, j, t)·τ(i, j, t) determines the state exchange path configuration;

[0091] For the update of the spatial gradient gating parameter matrix, the historical change gating parameter matrix and the mixing ratio parameter α, a probabilistic perturbation strategy under the control of the lattice cooling mechanism is introduced: that is, according to the current temperature variable T kControl the perturbation amplitude of each parameter position Δ=∈·exp(-γT k ), where ∈ is the initial range of perturbation and γ is the cooling coefficient, which realizes the stable convergence of network structure and parameter space and compatibility with jump search;

[0092] S34. Record the structural configuration with the best fitness function value in all iteration rounds as the optimal structural configuration, which includes the optimized structural path tensor, spatial gradient gating parameter matrix, historical change gating parameter matrix and mixing ratio parameter, and apply the optimal structural configuration to the neural switch network.

[0093] The present invention uses the ice crystal ant colony optimization algorithm to jointly optimize the structural configuration and control parameters in the neural switch network, adopts a path tensor adjustment mechanism based on structural heuristic factors and cross-pheromones, and combines it with a gated perturbation update method under lattice cooling control. Compared with traditional neural structure search methods, it has stronger global convergence and parameter stability.

[0094] In this embodiment, the S4 specifically includes:

[0095] S41, deploying the optimal structural configuration obtained by the ice crystal ant colony optimization algorithm to the neural switch network to form an optimized network with a fixed structure;

[0096] S42, inputting the constructed pressure map sequence into the optimization network, and the network sequentially extracting the node state as the initial activation state at each time step;

[0097] S43. In each time step, based on the state exchange path defined in the structural path tensor, the state relationship between the nodes is judged; when a node meets the activation conditions of its spatial gradient gate value and historical change gate value, and the value corresponding to the node in the current time step and the other node in the previous time step in the structural path tensor is 1, the current state is fused with the state of the associated node in the previous time step according to the mixing ratio parameter corresponding to the node;

[0098] S44. After the fusion process is completed, the replaced node state is saved as a historical version, the original state of the node in the previous time step is recorded as a memory vector, and the states of all nodes that have undergone state exchange and completed fusion in the current time step are combined with the corresponding memory vectors to form intermediate control information;

[0099] S45. Repeat the above state exchange and intermediate control information construction process until all time steps are traversed and a complete intermediate control information sequence is finally obtained.

[0100] The present invention performs dynamic state transfer and memory generation based on the optimized path structure, gating module and fusion mechanism in the neural switch network, effectively reconstructs the timing pattern in the pressure map sequence and extracts intermediate control information. Compared with the static network model, it has more real-time and context understanding capabilities, and significantly enhances the processing capability of continuous state input.

[0101] In this embodiment, the S5 specifically includes: reading the intermediate control information output by the neural switch network, extracting the fusion state of each node and the corresponding memory vector, combining them in chronological order to form the input data required for adjustment, dividing the input data into control components corresponding to each receiving cavity adjustment unit, and converting them into data parameters indicating the action amplitude and execution timing of each adjustment unit, inputting the data parameters into the drive interface of the receiving cavity actuator to complete the structural deformation.

[0102] The present invention constructs an adjustment parameter set based on intermediate control information and drives the receiving cavity actuator to complete structural deformation, avoiding response hysteresis and error diffusion based on fixed rule mapping, establishing a closed-loop path of perception-decision-execution, and improving the immediate response capability of the prosthetic system and user wear adaptability.

[0103] Example 1:

[0104] In order to verify the feasibility and technical advantages of the present invention during implementation, the pressure-sensing-based prosthetic socket automatic adjustment system proposed in the present invention was applied to a lower limb prosthesis adaptation scenario, and the system was deployed and field-tested on a number of users who wear prostheses for a long time. During the test, a high-resolution thin-film pressure sensor array continuously collected a series of socket contact pressure diagrams of the user under various typical actions such as standing, walking, and going up and down slopes. The neural exchanger network was used for time series modeling, and the ice crystal ant colony optimization algorithm was used to complete the joint optimization of the network structure and control parameters. The optimized neural exchanger network can automatically generate intermediate control information, drive the flexible deformation mechanism to adjust the internal structure of the socket, and thus realize the automatic adjustment of the socket under different action states.

[0105] During the system application process, in order to ensure the validity of the data comparison, the comprehensive performance of the traditional static socket structure, the adjustment system based on simple threshold feedback and the system proposed by this invention were compared. In the dynamic load-bearing gait test, the high-pressure area of ​​the traditional system during the load-bearing process was an average of 23.8cm 2 , and the high pressure area after adjustment by the system of the present invention is reduced to 7.1cm 2The average adjustment response time for 50 action cycles was 2.43 seconds for the traditional adjustment system, while the present invention only took 0.97 seconds, an improvement of more than 2.5 times in response efficiency. Furthermore, using a surface skin temperature thermal map to detect the distribution of local temperature rise within the receiving cavity, the mean hotspot distribution in the present invention system decreased by 12.3%, effectively reducing discomfort caused by prolonged wear.

[0106] In addition, in the adjusted user subjective comfort score, a 0-10 scale was used to evaluate the adaptability of each system to different motion states. The average score of the traditional system was 5.6, the system based on simple feedback scored 6.9, and the average score of the system of the present invention reached 8.5, significantly improving the user experience. In terms of system operation stability, the mean time between failures (MTBF) of the system of the present invention during the test was 127 hours, much higher than the 82 hours of the comparison system, reflecting strong system reliability. In order to intuitively demonstrate the data performance of the system of the present invention in actual application, the following test data table is compiled:

[0107] Table 1: Performance comparison between the system of the present invention and the prior art system

[0108]

[0109] From Table 1, "Performance Comparison of the Present Invention System and the Prior Art System," it is clear that the present invention outperforms traditional static socket structures and simple feedback adjustment systems in several key performance indicators. In terms of average high-pressure contact area, the present invention system effectively controls the local high-pressure area within the prosthetic socket to 7.1 cm. 2 , much lower than the 23.8cm of the static structure 2 and 15.4cm of simple feedback system 2 , indicating that it has higher adjustment accuracy and can better adapt to the dynamic pressure distribution of the user's residual limb.

[0110] In terms of response speed, the system's average adjustment response time is 0.97 seconds, significantly faster than the 2.43 seconds of a static structure with no adjustment and the 1.65 seconds of a feedback system. This indicates that the system can more quickly complete structural adaptive adjustments after user action, improving real-time wearing comfort. Furthermore, looking at the changes in the distribution of local temperature rise hotspots, the present invention shows that the average temperature distribution in the hotspot area decreases by 12.3% after dynamic adjustment, significantly higher than the 6.5% of the feedback system. The static structure has no adjustment function, so the average temperature decreases by 0%. This demonstrates that the system has a better heat regulation effect and can effectively reduce the risk of skin damage caused by concentrated pressure.

[0111] User subjective fit scores reflect the wearer's intuitive experience of the socket's comfort. The system's average score reached 8.5, surpassing the feedback system's 6.9 and the static structure's 5.6, demonstrating that the system provides a more ergonomic adjustment experience in practical applications. Finally, in terms of system reliability, the mean time between failures (MTBF) reached 127 hours, surpassing the feedback system's 105 hours and the static structure's 82 hours, demonstrating the stability and reliability advantages of the system in long-term operation.

[0112] In summary, the present invention significantly improves the adjustment accuracy, response speed, thermal comfort, user satisfaction and system stability of the prosthetic socket system by introducing a dynamic adjustment mechanism constructed by a neural exchanger network and an ice crystal ant colony optimization algorithm, and has obvious practical application value and industrialization prospects.

[0113] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A pressure-sensing based automatic adjustment system for prosthetic sockets, characterized in that: include: A pressure acquisition module is used to collect contact pressure data in the prosthetic socket and construct a pressure map sequence; The neural switch network modeling module is used to construct a neural switch network composed of several switching units. Each switching unit is equipped with a spatial gradient gating module and a historical change gating module, and defines the mixing ratio parameters and the switching path structure. a structure and parameter joint optimization module, for jointly optimizing the switching path structure, gating module, and mixing ratio parameters in the neural switch network using an ice crystal ant colony optimization algorithm to generate an optimized network structure configuration; The state processing and intermediate control information generation module is used to perform state exchange based on the optimized network structure configuration, perform state fusion through the mixing ratio parameter, generate memory vectors, and form intermediate control information; The adjustment instruction generation and driving module is used to convert the intermediate control information and the memory vector into adjustment input data, and further convert it into an execution control signal to drive the receiving cavity actuator to complete the structural deformation; The feedback acquisition and loop control module is used to collect contact pressure data after structural deformation, form an updated pressure map sequence and input it into the neural exchanger network to execute the next round of structural adjustment process.

2. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 1, characterized in that: The modules are implemented as follows: S1, collecting contact pressure data in the prosthetic socket and constructing a pressure map sequence; S2. Constructing a neural switch network, wherein the neural switch network includes a plurality of switching units, each of which is provided with two gating modules, a spatial gradient gating module and a historical change gating module, for controlling whether the node state participates in the exchange, setting a mixing ratio parameter for fusing the exchanged state with the original state, and defining a switching path structure for transferring the activation state between time steps; S3, using the ice crystal ant colony optimization algorithm to jointly optimize the switching path structure, gating module and mixing ratio parameters to obtain an optimized network structure configuration; S4. Applying the optimized network structure configuration to the neural switch network to obtain an optimized network, inputting the pressure map sequence into the optimized network, performing state exchange between nodes that meet the gating conditions according to the exchange path structure, using the mixing ratio parameter to perform state fusion, and introducing the replaced state of the previous time step to form a memory vector to generate an intermediate control information sequence for control; S5. Generate a receiving cavity adjustment instruction according to the intermediate control information sequence, and drive the actuator to generate structural deformation; S6. Collect contact pressure data after structural deformation, construct an updated pressure map sequence, and input it into the optimization network to execute the next round of structural adjustment process.

3. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 2, characterized in that: The spatial gradient gating module and the historical change gating module of each switching unit in the neural switch network of step S2 are used to jointly control whether the node state participates in the activation state exchange, specifically including: Get the node i and its adjacent node set at the current time step t The pressure value p t (i) and p t (j), where Calculate the spatial gradient gate value G of node i s (i,t); In the historical change gating module, the pressure value of node i in the past K time steps {p t-k (i)}, k=1,2,…,K, calculate the historical change gate value G h (i,t); The spatial gradient gating value and the historical change gating value are the outputs of the gating function that changes with time. The spatial gradient gating parameter matrix and the historical change gating parameter matrix in the network are calculated. G s (i,t) and G h (i, t) and the set threshold value θ s and θ h Compare, if and only if G s (i,t)>θ s And G h (i,t)>θ h When both are true, the activation state of node i participates in the exchange operation.

4. The pressure-sensing-based prosthetic socket automatic adjustment system according to claim 3, characterized in that: Each switching unit in the neural switch network of step S2 sets a mixing ratio parameter for fusing the original activation state of the node at the current time step with the switching state passed in from the previous time step, specifically including: At time step t, for node i that meets the exchange condition, let the original activation state be h t (i), the corresponding exchange state is the activation state h introduced from node j at time step t-1 t-1 (j), where And the gating condition is satisfied, that is, if and only if G s (i,t)>θ s And G h (i,t)>θ h hour; Assign a mixing ratio parameter α(i,t)∈[0,1] to node i to control the activation state fusion ratio; Based on the mixing ratio parameters, calculate the fusion state of node i at time step t 5. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 4, characterized in that: The switching path structure in the neural switch network of step S2 is used to specify the transmission mode of the activation state between different time steps, specifically including: in time step t, constructing a three-dimensional structure path tensor W swap ∈{0,1}, where W swap [i,j,t]=1 means that node i receives the activation state from node j at time step t-1 at time step t. If W swap [i,j,t]=0 means that node i does not receive the activation state from node j at time step t-1 at time step t. For each node i that meets the gating condition, according to the adjacent node set With the structure path tensor, select the only exchange node Determine the exchange path weight ω(i,j,t)∈{0,1} so that Get the activation state h of node j at time step t-1 from the structure path tensor t-1 (j), and combined with time step t node h t (i), call the fusion state The fused state is output as the update of node i at the current time step.

6. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 5, characterized in that: The S3 specifically includes: S31, constructing an ant colony population set, where each ant individual represents a structural configuration, wherein the structural configuration includes a structural path tensor, a spatial gradient gating parameter matrix, a historical change gating parameter matrix, and a mixing ratio parameter; S32, dividing the constructed pressure map sequence into a training set and a validation set, fitting the training set using the neural switch network under the current structural configuration, and performing the control information output process on the validation set, setting the fitness function as the difference score between the intermediate control information generated by the neural switch network on the validation set and the target control configuration, using the Manhattan distance function to compare the two, and using the score as an evaluation criterion for the quality of the structural configuration; S33. In each round of iteration, the following optimization process is performed: For the structure path tensor W swap The update introduces the structural inspiration factor η s (i, j, t) and the cross pheromone strength τ(i, j, t), by calculating the probability weight P(i, j, t) ∝ η s (i, j, t)·τ(i, j, t) determines the state exchange path configuration; For the update of the spatial gradient gating parameter matrix, the historical change gating parameter matrix and the mixing ratio parameter α, a probabilistic perturbation strategy under the control of the lattice cooling mechanism is introduced: that is, according to the current temperature variable T k Control the perturbation amplitude of each parameter position Δ=∈·exp(-γT k ), where ∈ is the initial range of perturbation and γ is the cooling coefficient, which realizes the stable convergence of network structure and parameter space and compatibility with jump search; S34. Record the structural configuration with the best fitness function value in all iteration rounds as the optimal structural configuration, which includes the optimized structural path tensor, spatial gradient gating parameter matrix, historical change gating parameter matrix and mixing ratio parameter, and apply the optimal structural configuration to the neural switch network.

7. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 6, characterized in that: The S4 specifically includes: S41, deploying the optimal structural configuration obtained by the ice crystal ant colony optimization algorithm to the neural switch network to form an optimized network with a fixed structure; S42, inputting the constructed pressure map sequence into the optimization network, and the network sequentially extracting the node state as the initial activation state at each time step; S43. In each time step, based on the state exchange path defined in the structural path tensor, the state relationship between the nodes is judged; when a node meets the activation conditions of its spatial gradient gate value and historical change gate value, and the value corresponding to the node in the current time step and the other node in the previous time step in the structural path tensor is 1, the current state is fused with the state of the associated node in the previous time step according to the mixing ratio parameter corresponding to the node; S44. After the fusion process is completed, the replaced node state is saved as a historical version, the original state of the node in the previous time step is recorded as a memory vector, and the states of all nodes that have undergone state exchange and completed fusion in the current time step are combined with the corresponding memory vectors to form intermediate control information; S45. Repeat the above state exchange and intermediate control information construction process until all time steps are traversed and a complete intermediate control information sequence is finally obtained.

8. The pressure-sensing-based automatic adjustment system for prosthetic sockets according to claim 7, characterized in that: The S5 specifically includes: reading the intermediate control information output by the neural switch network, extracting the fusion state of each node and the corresponding memory vector, combining them in chronological order to form the input data required for adjustment, dividing the input data into control components corresponding to each receiving cavity adjustment unit, and converting them into data parameters indicating the action amplitude and execution timing of each adjustment unit, inputting the data parameters into the drive interface of the receiving cavity actuator to complete the structural deformation.

Images