WO2021221672A1

WO2021221672A1 - Customized parametric models to manufacture devices

Info

Publication number: WO2021221672A1
Application number: PCT/US2020/030802
Authority: WO
Inventors: Vijaykumar Nayak
Original assignee: Hewlett-Packard Development Company, L.P.
Priority date: 2020-04-30
Filing date: 2020-04-30
Publication date: 2021-11-04

Abstract

An example non-transitory computer-readable medium includes instructions to receive a measurement dataset that is based on a measurement of a users body, apply a transformation to the measurement dataset to obtain a transformed measurement dataset in the domain of a parametric model of a device to be worn by the user, adjust the parametric model based on the transformed measurement dataset to obtain a customized model of the device that is customized to the user, output the customized model to enable manufacture the device, and adjust the transformation based on a feedback dataset to obtain an adjusted transformation and to apply the adjusted transformation to generate a subsequent customized model to enable manufacture a subsequent device.

Description

CUSTOMIZED PARAMETRIC MODELS TO MANUFACTURE DEVICES

BACKGROUND

[0001] Various types of devices, such as orthoses, prostheses, protective equipment, and equipment used for certain activities, may be worn by users to provide a benefit. Such devices may be used by many people as treatment for physical conditions or ailments, to improve athletic performance, and to improve quality of life. Examples of such devices include footwear inserts or insoles, specialized footwear, helmets, artificial limbs, and various types of braces. For example, a person may insert an insole into their shoe to correct gait and stance and potentially improve comfort and quality of life.

BRIEF DESCRIPTION OF THE DRAWINGS

[0002] FIG. 1 is a block diagram of example instructions to generate a customized model of a worn device and adjust the process of device generation based on received feedback.

[0003] FIG. 2A is a schematic diagram of example partially overlapping model and measurement domains and an example non-overlapping feedback domain.

[0004] FIG. 2B is a schematic diagram of example coincident measurement and feedback domains and an example non-overlapping model domain.

[0005] FIG. 2C is a schematic diagram of example partially overlapping model, measurement, and feedback domains. [0006] FIG. 3 is a flowchart of an example method to manufacture a customized worn device with adjustment to device models based on received feedback.

[0007] FIG. 4 is a flowchart of an example method to manufacture a customized worn device with feedback to adjust a matrix transformation used to transform captured data to model data.

[0008] FIG. 5 is a block diagram of an example system to generate a customized model of worn device and adjust model generation based on received feedback.

DETAILED DESCRIPTION

[0009] It is often the case that worn devices are designed for general use for a group of people that share a common need. Customization is frequently a laborious process that may involve much individualized professional expertise. Learnings are often not shared and customization techniques that are found to work well for one individual are not readily transported to the situations of others and the industry as a whole. People who rely on these devices may see mixed results that depend largely on the skills and experience of the professionals assisting them.

[0010] A worn device, such an orthotic device, a prosthetic device, a protective device, or a specialized device worn for a specific activity may be designed by capturing data from the body of a person who will be using the device. Data of the relevant portion of the person’s body may be captured. The data may be transformed to be in a domain of a parametric model of the device. The transformation takes user-specific measurement data, which does not necessarily correspond to the geometry and material of the device, and transforms the user-specific measurement data so that it may be readily applied to a parametric model of the device to obtain a personalized model. Hence, the parametric model or template may be customized based on user-specific data that need not directly align to the parameters that define the model. [0011] The user-specific customized device may be manufactured, for example, by three-dimensional (3D) printing.

[0012] Feedback from the user or from other users may be used to fine-tune the customization process. The transformation may be adjusted based on feedback from the user after using their customized device or based on feedback from other users using their own customized devices. The transformation is improved with disparate feedback rather than merely direct feedback from one user about their own device. Moreover, the feedback may be of different character from the original measured data and may include qualitative feedback such as an evaluation of how the device feels and functions.

[0013] Examples of suitable worn devices are given throughout this disclosure and these examples are not intended to be limiting. Suitable worn devices include devices that tend to rely on biometric or user-specific measurement data, can be customized to the user’s body based on such data, and allow for user feedback concerning the performance, fit, comfort, usefulness, and other properties of the device in use. Orthotic and prosthetic devices are examples of suitable worn devices. The terms “orthotic” and “prosthetic” are used herein to denote devices that may be worn to provide an orthotic, orthopedic, or prosthetic function, such as treatment, correction, performance enhancement, comfort, and quality of life.

[0014] Example worn devices include footwear insoles, footwear midsoles (i.e., the component between the upper and lower), orthotic inserts, artificial limbs, braces, helmets, mouthguards, pads (e.g., kneepads, shoulder pads, etc.), bulletproof vests, body armor, and similar devices whose performance improves when customized to individual body sizes, proportions, and kinematics/dynamics and when customized to individual needs.

[0015] FIG. 1 shows an example non-transitory computer-readable medium 100 that includes instructions 102 to generate a customized model of a worn device 104, such as an orthotic or prosthetic device, and to adjust the process of device generation based on received feedback. The device 104 is customizable to an individual user, as will be discussed herein.

[0016] The device 104 is defined by a parametric model 106. The parametric model 106 is a generic model that includes parameters that may be modified to generate a customized model 108. Examples of parametric models 106 include computer-aided design (CAD) models defined in terms of geometry and material properties (e.g., material identifier, density, etc.). Parametric modelling techniques such as constructive solid geometry, primitive instancing, spatial occupancy enumeration, and boundary representation may be used. The setting of a parameters or parameters of the parametric model 106 changes the model from generic to a customized model 108.

[0017] The parametric model 106 may be created with input from experts and professionals knowledgeable in the type of device 104. The techniques discussed in this disclosure consider expert and professional input, and replace or supplement such input with automated processing of user-specific measurement and feedback. In this example, expert/professional input may be considered a reasonable starting point, and the parametric model 106 may thus be considered a generalized or template device that is not expected to function property for a specific individual without the automated data-driven customization discussed herein.

[0018] The parametric model 106 is configurable to fit a number of users, such as a group of people with a particular ailment, condition, or need. That is, properties of the parametric model 106 may be modified within the constraints of the model 106 to conform to the needs of a particular user 110 that will wear a customized device 116 manufactured according to the customized model 108.

[0019] Example parameters include parameters that define dimensions (e.g., distance, orientation, diameter, curvature, etc.), materials, the inclusion or omission of subcomponents, positions and orientations of included subcomponents, control points, anchor points, and similar properties. In the example of a footwear insole 104, a parametric model 106 may include parameters that define an outer boundary curve 112 of the insole 104 and parameters the define thicknesses 114 of the insole at various locations. During customization, a thickness parameter near the arch of the insole may be set based on the condition of a person’s arch, for example.

[0020] A parameter of the parametric model 106 may have an upper bound, a lower bound, or both. A constraint may be placed on an inputted parameter. A constraint may be placed on a dependent value, such as a resultant deformation or displacement at a location distant from a dimension that may be set. Constraints may be defined with the parametric model 106 and based on input from experts and professionals.

[0021] The instructions 102 configure the parametric model 106 to an individual user 110. The instructions 102 are executable by a processor or may be interpreted to generate instructions that are executable by a processor. The processor may include a central processing unit (CPU), a microcontroller, a microprocessor, a processing core, a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), or a similar device capable of executing instructions. The processor may cooperate with the non-transitory computer-readable medium 100, which may include an electronic, magnetic, optical, or other physical storage device that encodes instructions 102. The computer-readable medium may include, for example, random access memory (RAM), read-only memory (ROM), electrically-erasable programmable read-only memory (EEPROM), flash memory, a storage drive, an optical device, or similar.

[0022] The instructions 102 receive a measurement dataset 120 that is based on a measurement 122 of a user’s body 110. The instructions 102 may control a capture device to capture the measurement 122. The instructions 102 may receive the measurement 122 from a capture device that is controlled in another way.

[0023] The measurement dataset 120 may include biometric data captured from the user’s body. The term “body” as used herein is not limited to the entire body and may denote a body part, a portion of a body part, a joint, an appendage, and so on. Biometric data need not be directly indicative of physical anatomy and may include data related to motion, comfort, and similar.

[0024] The measurement dataset 120 may indude any number and type of measurements 122. A measurement 122 may be a direct measurement of at least a portion of a body part, for example, foot length. A direct measurement 122 may directly map to a parameter in the parametric model 106 and therefore be within the domain of the model 106. For example, a parametric model 106 of an insole may indude an overall length parameter that may be set based on a foot length measurement.

[0025] The measurement 122 may be indirect or outside the domain of the parametric model 106. For example, a measurement 122 may be related to a motion of the person’s body related to the customized device 116, such as the gait of the person 110 in the example of footwear, arm range of motion in the example of an elbow pad, leg motion in the example of a knee brace, and so on. Example gait measurements 122 indude a gait cyde, a step length, a step width, a step angle, a stance duration, a swing duration, a duration of double support, a duration of single support, and so on. Such a measurement 122 does not directly map to a parameter in the parametric model 106. For example, a thickness 114 of an example insole 104 cannot be set to any of the above- mentioned gait measurements.

[0026] The instructions 102 apply a transformation 124 to the measurement dataset 120 to obtain a transformed measurement dataset 126 that is within the domain of the parametric model 106. The transformation 124 may be linear. For example, the measurement dataset 120 may be formed as a matrix B. A transformation matrix T may be applied to the matrix B by matrix multiplication as follows:

[0027] A = TB

[0028] to obtain a matrix A containing the transformed measurement dataset 126. [0029] In another example, the transformation 124 may include a lookup table and thereby may be non-linear. Elements of the transformed measurement dataset 126 may be obtained by looking up measurements of the measurement dataset 120 in the lookup table. Interpolation may be used for values not literally in the lookup table.

[0030] In another example, the transformation 124 may include a trained machine-learning system, such as a trained neural network. The trained machine-learning system may take the measurement dataset 120 as input and provide the transformed measurement dataset 126 as output. The trained machine-learning system may be trained with historic designs and expert or professionally created designs.

[0031] The instructions 102 adjust the parametric model 106 based on the transformed measurement dataset 126 to obtain a customized model 108. An element of the transformed measurement dataset 126 may directly correspond to modifiable or settable parameter of the parametric model 106. Such an element may be directly applied to the model 106. As such, a customized model 108 of the device 104 may be obtained for the particular user 110 that provided the measurement dataset 120.

[0032] The instructions 102 may output the customized model 108 to enable manufacture of the customized device 116. The customized model 108 may be converted from a CAD format, such as IGES, DXF, DWG, STEP, ACIS, or STL, to a suitable 3D printing format, such as 3MF, STL, OBJ, or VRML. The customized model 108 may be provided to manufacturing apparatus, such as a 3D printer.

[0033] Manufacture by 3D printing may be used to fuse material, such as powder, to form a printed customized device 116. In a suitable powder-bed material fusion printing system, layers of powder are progressively introduced and select portions of each layer are fused with the previous layer. Material fusion may be performed using an energy source, a light source, laser, electron beam, a chemical fusing agent, binding agent, curing agent, an energy absorbing fusing agent, or combination of such that may be jetted or sprayed (e.g., via a thermal or piezo inkjet-type printhead), or similar. Fused layers thereby form a printed device and unfused material may be recovered and recycled.

[0034] The person 110 may then wear the customized device 116 and use it for its intended purpose.

[0035] Subsequently, the instructions 102 receive a feedback dataset 130 related to a customized device 116.

[0036] The instructions 102 use the feedback dataset 130 to apply an adjustment 132 to the transformation 124 to obtain an adjusted transformation 124. The adjusted transformation 124 may then be used to generate subsequent customized models 108 to enable the manufacture of subsequent customized devices 116 or the same or similar type.

[0037] The adjustment 132 may be determined from a comparison of the feedback data 130 to a reference standard. Taking gait for example, the feedback data 103 may include gait data captured when the person 110 is wearing the customized device 116. The gait data may be compared to reference gait data and a deviation from the reference gait data may be used to modify the transformation 124. A trained machine-learning system may be used to compare the feedback data 130 to the reference standard. The reference standard may represent an ideal gait and may be developed with the help of experts and professionals in the field. An mentioned, expert/professional input may provide a basis for the automated operations discussed herein, which supplement or replace human subjectivity. Moreover, expert/professional input may be used initially, such as to help create the generic model 106 or the feedback reference standard, however the automated operations discussed herein do not require continuous expert/professional input to provide ongoing customization of user-wearable devices. [0038] The adjustment 132 may be made irrespective of the source of the feedback dataset 130 and irrespective of any relationship of the feedback dataset 130 to previously made customized devices 116. The feedback dataset 130 may relate to any customized device 116 previously made and used by a person 110. That is, the adjustment 132 need not be specific to an improvement to a particular person’s customized device 116 so that, for example, the device may be remade. The adjustment 132 may thus generalize feedback to adjust the transformation 124 for manufacture of future customized devices 116, irrespective of whether such new devices are replacements of previous devices or are entirely new devices for new users 110. As such, future users may benefit from a continually improving process, despite differences in the customized devices 116.

[0039] The feedback dataset 130 may contain the same or similar data as the measurement dataset 120. The feedback dataset 130 may contain data that is partially or entirely different from the measurement dataset 120.

[0040] For example, a particular individual 110 may provide foot measurements 122, such as overall foot dimensions, plantar pressure measurements, a 3D foot scan, or other biometric data, and have a customized insole 116 manufactured based on such. The person 110 may use the insole 116 for some time, and later provide feedback by way of a personal qualitative evaluation of the insole 116. The evaluation may indicate that the insole 116 feels too firm. Accordingly, an adjustment 132 may be made to the transformation 124, so that future customized inserts may be less firm. This adjustment 132 may apply to future insoles 116 made, whether for the same person 110 who provided the feedback or another person. It is noted that the adjustment 132 is not necessarily made to the customized model 108 for the particular individual 110 who provided the feedback, as this would only benefit this specific person. Rather, the adjustment 132 is made to the transformation 124 that transforms measured data for anyone into transformed data that may be used to generate customized models 108 for anyone. Hence, individual feedback may be used to improve the process for numerous future users. [0041] It is noted that in the above example, the feedback dataset 130 is entirely outside the domain of the measurement dataset 120. The qualitative feedback of “too firm" is different from foot measurements. Further, the measurement dataset 120 may have some overlap with the parameters of the parametric model 106, such as overall foot dimensions. FIG. 2A shows this example as partial overlap of a parametric model domain 200 and a measurement domain 202, as well as no overlap of the feedback domain 204 and the measurement domain 202.

[0042] In another example, gait measurements 122 may be taken to generate a customized midsole 116 for an article of footwear for a particular person 110. Such measurements 122 may be taken under the supervision of a health professional that is helping the person 110 correct their gait using customized orthopedic footwear. The person 110 may return for follow-up gait measurements using the same techniques as the initial measurements 122, and thus the feedback dataset 130 may belong to the same domain as the initial measurement dataset 120. Both datasets 120, 130 may contain gait measurements captured under the same protocol. An adjustment 132 may be made to the transformation 124 and a new customized midsole 116 may be created for a new article of footwear for the person, so that their gait may continue to be corrected. Further, since it is the transformation 124 that is adjusted, other users may also benefit from this user’s experience.

[0043] FIG. 2B shows example domain overlap of the above example. A parametric model domain 210 and measurement domain 212 do not have overlap, while a feedback domain 214 and the measurement domain 212 coincide.

[0044] FIG. 2C shows another example of domain overlap that indude partial overlap among a parametric model domain 220, a measurement domain 222, and a feedback domain 224. Each domain partially overlaps with both other domains. This may occur, for example, in a scenario in which a target firmness of an insole or midsole is initially measured and is used to set a material parameter of the parametric model (e.g., to select a material density), and an evaluation of the firmness of the manufacture device is received as feedback.

[0045] It is noted that a parametric model domain is quantitative, in that a parametric CAD model takes definite and often numeric input. Measurement and feedback domains may be quantitative or qualitative. Quantitative data may be captured using a sensor such as those discussed herein. Qualitative data may be captured by user input. Examples of qualitative data include questionnaires, ratings, and selections from choices. Qualitative data may be converted to numeric values using a mapping, so that a matrix operation or other type of transformation may be carried out.

[0046] FIG. 3 shows an example method 300 to manufacture a customized worn device with adjustment to device models based on received feedback. The method 300 may be implemented with instructions that may be stored in a non- transitory computer-readable medium and executed by a processor, as discussed elsewhere herein. The description provided elsewhere herein may be referenced for details not repeated here, with like reference numerals and like terminology denoting like components.

[0047] At block 302, data is captured for the creation of a customized instance of device, such as an orthotic or prosthetic device or other type of device discussed herein. Captured data may include a biometric measurement of the body of a person who will wear the device. For example, if the device is a footwear insole or midsole, captured data may include data concerning the foot and gait of the person.

[0048] Captured data may include information provided by the person, such as a preference or a description of an ailment or condition. For example, questionnaire may be administered. In the case of a footwear insole or midsole, data volunteered by the person may include a firmness preference, an indication of any discomfort when walking, an activity level, and similar information. [0049] Captured data may include information provided by a professional or expert treating the person for whom the device is to be made. For example, a health professional may interview or treat the person and obtain information, such as data related to diagnosis, assessment, or treatment. In another example, an athletic consultant may examine the person’s running or walking gait to collect data for the manufacture of a customized performance enhancing device.

[0050] At block 304, the captured data is transformed to the domain of a parametric model that defines the device generically. A transformation may be defined to map captured data in a first domain to transformed data in a second and different domain. The first domain includes measurement and other input data that may not directly correspond to the parameters that define the parametric model. The second domain includes the physical and material parameters of the parametric model and may have some or no overlap with the first domain.

[0051] The transformation may be linear or non-linear. An example of a linear transformation is a matrix operation that is performed on a matrix that contains the data of the first domain to obtain data of the second domain. Examples of a non-linear transformation include a lookup table, a set of functions, a trained machine-learning system, and similar.

[0052] The transformation may be initially defined based on historic data, empirical data, and/or professional/expert input. As cycles of the method 300 are performed, the transformation is modified based on feedback and thus gradually departs from its initial state.

[0053] At block 306, the transformed data is used to customize the parametric model of the device. This may include setting values of parameters of the model to values extracted from the transformed data. For example, the transformed data may be a matrix of values, where each value corresponds to a different parameter of the model. A customized model that is specific to the transformed data, and possibly unique, is obtained. [0054] Further, the customized model may be converted into a format expected by the manufacturing apparatus that is to be used. For example, the customized model may be converted from a CAD format to a suitable 3D printing format.

[0055] At block 308, the customized model in a suitable format is provided to the manufacturing apparatus and the apparatus makes the customized device.

If different apparatuses are used, each is provided with the relevant data from the customized model.

[0056] At a later time, at block 310, feedback related to the customized device, as worn and used by the user, may be received. Feedback may include data of a third domain, which may be entirely separate from, may overlap, or may be coincident with the first domain of measurement and other input data. For example, feedback data may be obtained by a follow-up with the professional overseeing treatment or therapy.

[0057] At block 312, the transformation used to transform data of the first domain to the second (model) domain is adjusted based on the feedback. The adjustment may include comparing the feedback data to reference data. For example, plantar pressure data (first domain) may be used to customize a parametric model of an insole (second domain) for a particular user, at blocks 302-308. Later, a gait analysis is performed on the user while wearing the insole. Obtained gait data (third domain) is taken as feedback and is compared to a reference gait. A deviation from the reference gait is used to adjust the transformation. For example, the gait data may indicate a heel strike, so an element of the transformation that customizes the relevant portion of the customized model may be adjusted to generate a new customized model that may reduce heel strike. Since it is the transformation that is adjusted, future customized models, for this user and other users, will inherently take the feedback into account.

[0058] Adjustment of the transformation may be facilitated by the transformation being transparent. For example, in the case of a matrix operation, the relationship between elements of a transform matrix T and parameters of the parametric model may be clear and definite. As such, adjustment to the transform matrix T may be performed by adjusting its elements that correspond to the parameters of the parametric model that are to be adjusted.

[0059] A machine-learning system used as the transformation may be opaque and the relationship between the inputs of the machine-learning system and outputted parameters of the parametric model may be unknown. Adjusting the transformation in this case may be done by modifying the customized model according to the comparison of feedback and reference data, and then using the modified customized model for additional training of the machine-learning system.

[0060] The method 300 may be repeated indefinitely, via block 314, to manufacture a plurality of customized devices for various individuals. The method 300 may increase the usability, performance, or efficacy of manufactured devices over time based on feedback. The feedback need not be of the same domain as the data used to manufacture the devices. Further, feedback related to an individual’s customized device can improve customized devices manufactured for other individuals, despite the fact that customized devices may be different and possibly unique.

[0061] FIG. 4 shows an example method 400 to manufacture a customized device with feedback to adjust a matrix transformation used to transform captured data to model data. The method 400 may be implemented with instructions that may be stored in a non-transitory computer-readable medium and executed by a processor, as discussed elsewhere herein. The description provided elsewhere herein may be referenced for details not repeated here, with like reference numerals and like terminology denoting like components.

[0062] At block 302, data, such as biometric data, is captured from a user 402. The data may be measured directly from the body of the user. Professional or expert input 404 may be applied to capture the data. For example, a health professional may oversee or guide the collection of data.

[0063] At block 406, the captured data is cleaned to remove noise, outliers, or spurious datapoints. Professional or expert input 404 may also assist with the cleaning of data.

[0064] Professional or expert input 404 may be direct human input or may be automatic input that is configured based on human input, such as expert guidelines. Professional or expert input 404 may be advisory or assistive in nature and does not directly influence the model transformation and feedback proceed. For example, a professional may instruct the user how to stand on a planar pressure sensor and help quantify the foot discomfort they feel.

[0065] At block 408, key features are extracted from the cleaned captured data. The captured data may be rich and descriptive, and feature extraction may be performed according to a suitable technique to obtain a relatively smaller set of salient features that represent the captured data. For example, the captured data may include an array of pressure sensor measurements. Feature extraction may identify regions of different pressure level ranges, a point or region of greatest pressure, a point or region of least pressure, a foot outline, a foot length or width, and so on. If the captured data includes gait data, then feature extraction may identify various durations, positions, and orientations. A specific feature extracting technique employed may depend on the type of data captured. Examples of feature detection include thresholding, blob extraction, min/max detection, and edge/contour detection.

[0066] Block 408 may further include identifying key dimensional or structural measurements, which in the example of insoles and footwear may include features of shape and size of different regions of pressure, an overall size of a foot (length and width), an arch position, an arch height, an arch shape, a gait cycle, a step length, a step width, a step angle, a stance duration, a swing duration, a duration of double support, a duration of single support, and so on. [0067] Block 408 may generate a matrix B containing values that define the key features of interest in the customization of the device. Matrix B may include dimensional information related to the body of the user 402 and this information may be in a different domain from dimensional parameters of a parametric model of the device. That is, the dimensional information in matrix B may not be readily inputted into the parametric model to obtain a customized model. Matrix B may further include other information, such as material information (e.g., material identifiers, densities, efc.), and this may or may not be in the same domain as the parametric model.

[0068] Matrix B may have one row of m values, such as:

[0069] B = [B₁ B₂ B₃ ... B_m]

[0070] At block 410, a transformation 416 may be selected from a library of transformations 412. The library may include transformations specific to different conditions of users and different types or purposes of customized devices. For example, a transformation for supination may be provided and a different transformation for pronation may also be provided. The transformation 416 may be selected based on the extracted features at block 408. Expert or professional input 404 may additionally or alteratively be used to select the transformation 416.

[0071] The transformation 416 may be a matrix T of coefficients of width n and height m, such as:

[0072]

[0073] At block 414, the matrix B that defines the features extracted from the captured data is transformed using the selected transformation 416 to obtain a transformed matrix A. Matrix multiplication may be performed, as follows:

[0074] A = TB [0075] The transformed matrix A may have one column of n values, such as:

[0076]

[0077] Matrix A may contain a set of values for parameters of a parametric model.

[0078] At block 418, a parametric model 424 may be selected from a library of parametric models 420. The parametric model 424 may be selected as a model that corresponds to the selected transformation 416. Libraries of parametric models 420 and transformations 412 may be constructed for different conditions of users and different types or purposes of customized devices. For example, a library of parametric models 420 may include generic models for various foot shapes and conditions, such as pronation, supination, plantar fasciitis, fallen arches, and so on. The library of transformations 412 may include different transformations that correspond to the parametric models 420. A one-to-one correspondence may be established, in that each model 420 has one associated transformation 412.

[0079] At block 422, the selected parametric model 424 is customized to the user 402. The matrix A of parameter values is applied to the selected parametric model 424. A value may be directly inputted into a parameter of the model 424. This customizes the selected parametric model 424 based on the user-specific matrix A. A customized model is thus obtained.

[0080] At block 426, the customized model is manufactured. The customized model may be 3D printed to obtain a customized device specific to the user 402.

[0081] After the customized device is made, it may be used for a duration. Feedback related to the use of the device by the user 428, is collected, at block 430. Feedback may include a qualitative evaluation by the user 402, such as a questionnaire or similar inquiry. Qualitative data can be mapped to quantitative values for further processing of the feedback (e.g., good = 3, fair = 2, bad = 1). Feedback may additionally or alternatively include data captured from the user with the customized device worn. Such captured data may be of the same or different domain to the captured data 302 used to manufacture the device. Expert or professional input 404 may be used to assist in capture of the feedback.

[0082] At block 432, key indicators are extracted from the feedback data.

The key indicators may correspond to a trained machine-learning system 434 used to implement adjustments based on the feedback. For example, a qualitative evaluation of a customized insole may indicate discomfort in a toe or heel region of the user’s foot. A key indicator may be extracted from this information as a value indicative of type of discomfort, location of discomfort, and degree of discomfort. In another example, feedback 430 may include gait analysis that provides a range of gait data of the user when wearing the customized device. Key indicators may be extracted from such an analysis, such as gait cycle, a step length, a step width, a step angle, a stance duration, a swing duration, a duration of double support, a duration of single support, and so on.

[0083] The feedback key indicators obtained at block 432 may correspond to like indicators in a reference standard 436. The feedback key indicators may be stored in a matrix M’ and the reference standard key indicators may be stored in a matrix M.

[0084] At block 438, an adjustment is made to the transformation 416 based on a comparison of the matrix M’ of the feedback key indicators to the matrix M of the reference standard 436. The adjustment may be made by a trained machine-learning system 434, such as a trained neural network.

[0085] The adjustment at block 438 may seek to reduce of minimize an error or deviation between the matrix M’ of the feedback key indicators and the matrix M of the reference standard, as may be expressed as follows: [0086] error = M — M'

[0087] The adjustment may be formed as a matrix ΔΤ of adjustment values that is proportional to the error. The adjustment matrix ΔΤ may be added to the transformation matrix T to arrive at an updated transformation 416 that is used for subsequent customization of the corresponding parametric model 424. The updated transformation 416 may replace its former version in the library of transformations 412.

[0088] Block 438 may also consider matrix A of model parameters for the user and device 428 providing the feedback. That is, matrix A may be provided as input to the trained machine-learning system 434. This provides context for the comparison of the user-specific feedback data of matrix M^* with the reference standard of matrix M. In a simplified but illustrative example, if matrix A describes a model of a small insole (/.e., a user with small feet), this influences the comparison of the matrices M and M’ to provide a transformation adjustment matrix ΔΤ with relatively small corrections. For a larger insole, corrections may be coarser to have a comparable effect In any case, matrix A provides information that can make the adjustment to the transformation more accurate, as matrix A carries information about the transformation described by matrix T and the user as described by matrix B.

[0089] In other examples, user-data matrix B and/or transformation matrix T may be provided to block 438 in addition to or instead of model-parameter matrix A.

[0090] The same applies to examples that use lookup tables, neural networks, or other type of transformation, at block 414. In addition to feedback and reference data, the adjustment at block 438 can take as input contextual about the user, the model, and/or the transformation.

[0091] The method 400 thus provides for customized design of individual devices, as well as feedback that contributes to improvements in future customized designs of such devices by way of adjustments to a transformation. [0092] FIG. 5 shows an example system 500 to generate a customized model of an device and adjust model generation based on received feedback. Features and aspects of the various devices, systems, and methods discussed herein may be used with the system 500. Details of similar components will not be repeated here and components with similar terminology or reference numerals may be referenced for further information.

[0093] The system 500 includes a data acquisition device 502, 504, a processor 506, memory 508, a user interface 510, and instructions 102. In this example, the components of the system 500 may be provided as a single unit, whether contained in a single housing or multiple connected housings, that may be installed at a particular location, such as a retail store, health clinic, doctor’s office, or the like. In other examples, various components may be situated at various locations and may communicate via a computer network.

[0094] Various individual and combinations of data acquisition devices 502, 504 may be provided.

[0095] A plantar data acquisition device 502 may include a pressure sensor array, sensor mesh, or similar device that captures a two-dimensional dataset of pressures 514 exerted on a platform or surface that is shaped and sized to be stood on by a person 110. The plantar data acquisition device 502 may include a capacitive sensor array. The plantar data acquisition device 502 may be communicatively connected to the processor 506 by an input/output circuit (I/O) 512, such as a serial connector, parallel connector, wireless adaptor, or similar.

[0096] An image acquisition device 504 may include a camera or other image sensor to capture images of a person 110 while walking or running. The image acquisition device 504 may be used to capture gait data. The image acquisition device 504 may be communicatively connected to the processor 506 by the I/O circuit 512.

[0097] The processor 506 may include a CPU, microcontroller, microprocessor, processing core, FPGA, ASIC, or a similar device capable of executing instructions. The memory 508 is connected to the processor 506 and may include RAM, ROM, flash memory, a storage drive, an optical device, or similar non-transitory computer-readable medium.

[0098] The user interface 510 is connected to the processor 506 and may include a display, speaker, keyboard, touchscreen, or similar device to allow a user or a professional/expert to operate the system 500.

[0099] The instructions 102 may be stored in the memory 508 and may control the data acquisition device 502, 504 to capture data 516 related to the body of the user 110. The instructions 102 may further select a transformation 518 for the captured data, apply the transformation 518 to obtain transformed data 520, select a parametric model 522, and apply the transformed data 520 to a selected parametric model 522 to obtain a customized model 524 of a device 116 for the user 100. The customized model 524 may be manufactured to provide a customized device 116 to the user 110.

[0100] Feedback data 526 may be captured with regard to the user’s 110 usage of the device 116. Feedback may be captured with the same or different data acquisition device 502, 504 used to manufacture the device 116. A user interface 510 may be used to provide other types of feedback, such as a qualitative evaluation of the device 116. The instructions 102 may apply feedback data to adjust the transformation 518, so that future devices are manufactured in accordance with the feedback.

[0101] A plurality of transformations 518 and parametric model 522 may be maintained for various worn devices suitable for various different conditions or needs. Multiple sets of transformed data 520, feedback data 526, and customized models 524 may be obtained for multiple different users 110.

[0102] In view of the above, it should be apparent that user-wom device may be customized from a parametric model using a transformation to transform data that is captured about the user of the device into the domain of the model. Feedback about use of the device may be applied to the transformation to enable continuous improvement to future customized devices. Human subjectivity in the device customization and feedback processes may reduced or eliminated, while still allowing expert/professional guidance for generic model design and biometric data capture.

[0103] It should be recognized that features and aspects of the various examples provided above can be combined into further examples that also fall within the scope of the present disclosure. In addition, the figures are not to scale and may have size and shape exaggerated for illustrative purposes.

Claims

1. A norvtransitory computer-readable medium comprising instructions to: receive a measurement dataset that is based on a measurement of a portion of a user’s body; apply a transformation to the measurement dataset to obtain a transformed measurement dataset in the domain of a parametric model of a device to be worn by the user; adjust the parametric model based on the transformed measurement dataset to obtain a customized model of the device that is customized to the user; output the customized model to enable manufacture the device; and adjust the transformation based on a feedback dataset to obtain an adjusted transformation and to apply the adjusted transformation to generate a subsequent customized model to enable manufacture a subsequent device.

2. The non-transitory computer-readable medium of claim 1 , wherein the feedback dataset is based on a follow-up measurement of at least a portion of the user’s body.

3. The non-transitory computer-readable medium of claim 1 , wherein the feedback dataset is based on a measurement of at least a portion of a different user's body.

4. The non-transitory computer-readable medium of claim 1 , wherein the feedback dataset includes a qualitative evaluation received from the user or another user.

5. The non-transitory computer-readable medium of claim 1 , wherein the instructions are further to: compare the feedback dataset to a reference dataset; adjust the transformation based on the comparison of the feedback dataset to a reference dataset

6. The non-transitory computer-readable medium of claim 1, wherein the instructions are further to: select the transformation as a first transformation or a second transformation that is different from the first transformation, wherein the first transformation corresponds to a first condition of the user’s body, and wherein the second transformation corresponds to a second condition of the user’s body that is different from the first condition.

7. The non-transitory computer-readable medium of claim 1, wherein: the measurement data includes a measurement captured in relation to a foot or gait of the user; the parametric model includes a three-dimensional model of a midsole or insole; and the device includes the midsole or insole.

8. A system comprising: memory to store a transformation; a processor connected to the memory, the processor to: apply the transformation to a dataset captured from a user to obtain a transformed dataset; customize a parametric model of an orthotic or prosthetic device for the user based on the transformed dataset to obtain a customized model of the orthotic or prosthetic device; output the customized model to enable manufacture of the orthotic or prosthetic device; and adjust the transformation based on feedback related to use of the orthotic or prosthetic device by the user.

9. The system of claim 8, wherein the processor is further to apply the transformation to the dataset using a linear matrix operation.

10. The system of claim 8, wherein the processor is further to adjust the transformation based on the feedback by applying a machine-learning system to the feedback and to a reference standard.

11. The system of claim 8, wherein the processor is to select the transformation and the parametric model from respective libraries based on a condition or need of the user.

12. The system of claim 8, further comprising a data acquisition device connected to the processor to capture the dataset from the user.

13. A method comprising: applying a transformation to a biometric dataset captured from a user to obtain a transformed dataset; configuring a parametric model of an orthotic or prosthetic device for the user based on the transformed dataset to obtain a user-specific model of the orthotic or prosthetic device; and adjust the transformation based on feedback related to use of the orthotic or prosthetic device by the user.

14. The method of claim 13, further comprising capturing the biometric dataset from the user’s body.

15. The method of claim 13, further comprising three-dimensional printing the user-specific model to obtain a user-specific model orthotic or prosthetic device.