Biomechanics R&D Engineer

🎯 Context and goals

• The shared objective across this experience was to support product decisions with biomechanical, physiological, mechanical, and sensory evidence. Depending on the project, the goal was either to compare usage conditions, identify a target product profile, build a robust protocol before running a study, or rank technical solutions through quantitative analyses and scientific summaries that could be used internally.

🧪 Work carried out

Projects

• Domyos
• Kalenji
• NewFeel
• Oxelo
• Quechua

🧭 Scientific framing and protocol

• Domyos: a preliminary study on an innovative elliptical bike prototype comparing three usage modes (Classic, Kangaroo, Twist) with joint energy-expenditure and EMG measurements. The observable protocol combines randomized conditions, standardized stages, and physiological instrumentation to objectify product effects before broader deployment. Source: RE-16-ACAP-005 — Domyos oral presentation — 10 November 2016

Study objectives
Compare "Kangaroo" and "Twist"
against classic use
on energy expenditure (EE) and muscle activity (EMG)

• Kalenji: the project documents a thorough literature review on foot strike patterns (FSP) paired with an audit of in-store Footscan data to assess the tool's relevance in a retail setting. The review draws on nearly 40 scientific articles covering the comparative biomechanics of the three main patterns (RFS, MFS, FFS), associated injury profiles, running economy, the influence of drop and midsole stiffness, the effect of speed and surface type, and the link between athlete expertise and FSP. The dual objective is to deepen scientific understanding of FSP and to evaluate the operational validity of Footscan in-store. Source: FSP Literature Review — Decathlon Sports Lab 2016 / RE-16-ACAP-004 — Footscan Results — 2016

Why Footscan?
Determine the user's foot type
And recommend suitable shoe ranges
Total: 28200

• NewFeel: this project combines two complementary workstreams. The first searches for an ideal cushioning profile for fast walking through mechanical measurements and sensory evaluations; the second builds a fundamental description of foot roll-over by synchronizing plantar pressure, force, and kinematics. The protocol shows a full R&D logic, from sample selection to walk/run comparison and representative-subject identification. Source: ACAP-16-008 — Fundamental foot roll-over study — 2016/2017

The analysis covers both kinematics,
ground reaction forces and plantar pressures.
Compare variables in walking and running conditions

• Oxelo: this folder corresponds to a study-framing and methodological prototyping project for scooter evaluation, with no final experimental results in the available artifacts. The demonstrated value lies in protocol construction, selection of relevant biomechanical variables, and product-parameter exploration. Source: Oxelo study framing presentation — 2016/2017

Population: 21 male/female participants
Conditions: 10 scooter configurations
Observed variables: joint angles
Time to complete the slalom; perceived comfort

• Quechua: the project combines literature review, study proposal, mechanical testing, and statistical analyses to evaluate ankle support and select contrasted shoe models before a larger biomechanical study. The observable mechanical protocol is designed to objectify inversion speed and amplitude against a barefoot reference condition. Source: RE-ACAP-16-006 — Quechua ankle-support mechanical test results — 2016

Performance in terms of inversion speed and amplitude
is assessed on a 12-shoe bench to select the 5 most "different" pairs

📊 Data analysis

• Domyos: the Matlab scripts show a full processing chain for physiological and EMG data. Multi-subject import of Fitmate text files, decimal normalization, time reconstruction, condition-change detection, moving-average and low-pass smoothing, then calculation of interpretable indicators such as %heart-rate reserve, energy expenditure, oxygen cost, and caloric cost are all represented. The EMG pipeline follows a credible biomechanics workflow for product comparison: band-pass filtering, rectification, envelope extraction, area-under-the-curve computation, normalization against a reference condition, and removal of non-compliant trials. This approach demonstrates the ability to turn heterogeneous raw signals into synthetic variables that can be compared across trials and participants. Source: RunningEconomy_TestComparaisonPhysio_Raw.m — 2016

%% Weighted mean of parameters used in running economy
for j = 1:size(Vitesse,2)
    TempsEco{1,j} = Temps{1,j}(Index{1,j}(1):Index{2,j}(1)) - Temps{1,j}(Index{1,j}(1));
    for i = 1:(Index{2,j}(1)-Index{1,j}(1))
        Temp_Respi{1,j}(i+1,1) = TempsEco{1,j}(i+1)-TempsEco{1,j}(i);
    end
    VO2Eco_pond(1,j) = sum(VO2{1,j}(Index{1,j}(1):Index{2,j}(1)).*(Temp_Respi{1,j}(1:length(Temp_Respi{1,j}))/sum(Temp_Respi{1,j})));
    EEmEco_pond(1,j) = sum(EEm{1,j}(Index{1,j}(1):Index{2,j}(1)).*(Temp_Respi{1,j}(1:length(Temp_Respi{1,j}))/sum(Temp_Respi{1,j})));
end
VitesseEco_pond = VitesseEco_pond / 60;
Eq_cal = EEmEco_pond ./ (VO2Eco_pond/1000);
O2_cost1(i) = VO2Eco_pond(i) / (VitesseEco_pond(i) * masse);
CAL_cost_Fletcher1(i) = (VO2Eco_pond(i)/1000) .* Eq_cal(i) * K / (VitesseEco_pond(i) * 1000 * masse);

Source: Script_test_EMG2.m — 2016

Data1.(['Velo' num2str(j)]){n,k}.EMG.Filt.(param1{i}) = lpfilter(Data1.(['Velo' num2str(j)]){n,k}.EMG.Raw.(param1{i}),Fc2,fe,'damped');
Data1.(['Velo' num2str(j)]){n,k}.EMG.Filt.(param1{i}) = hpfilter(Data1.(['Velo' num2str(j)]){n,k}.EMG.Raw.(param1{i}),Fc1,fe,'damped');
Data1.(['Velo' num2str(j)]){n,k}.EMG.Rectified2.(param1{i}) = abs(Data1.(['Velo' num2str(j)]){n,k}.EMG.Filt.(param1{i}));

• Kalenji: the literature review draws on 5 reference studies for FSP distribution in running (Kerr 1983, Hasegawa 2007, Larson 2011, Kasmer 2013, Bertelsen 2013), covering cohorts of 248 to 1991 runners observed via high-frequency video at 10 km, half-marathon, and marathon distances. The analysis covers biomechanical differences between patterns (ground reaction forces, joint angles, muscle activity), injury profiles by FSP, controversies on running economy (including the limits of studies favoring FFS), the influence of drop and midsole stiffness, the speed-FSP relationship, and the link between athlete expertise and FSP. Nearly 40 articles are mobilized in total. On the field side, the Footscan dataset represents 28,200 scans; its analysis identifies protocol biases specific to a retail context and defines the minimum conditions for reliable use. Source: FSP Literature Review — 2016 / RE-16-ACAP-004 — Footscan Results — 2016

• NewFeel: this is the richest workstream in terms of data-processing automation. The Matlab scripts first read Footscan files step by step, separate right and left foot data, then synchronize signals acquired at 125 Hz, 250 Hz, and 2500 Hz. The pipeline continues with temporal interpolation, pressure-image rotation and resizing, computation of max/mean pressure and impulse, COP trajectories, six-zone segmentation, intra-foot and inter-trial averaging, then figure, poster, video, and export generation. The Dynamisme branch also builds an ANOVA-ready table for downstream analysis outside Matlab. This combination reflects strong biomechanics data engineering, multi-sensor alignment, and product-oriented analytical reporting. Source: RE_17_ACAP_008_Etude_du_deroule_du_pied_Pressions.m — 2017

Vtn  = 0:1/2500:(length(y1)-1)/2500;
Vtn2 = 0:1/125:(length(y2)-1)/125;
Var_Interp.([Name_Cond{k}]).Pressions.(Foot{k}{f}){o,1}(i,j,:) = interp1(Vtn2,squeeze(y2),Vtn);
Var_Interp.([Name_Cond{k}]).ForceZ.(Foot{k}{f}){o,1} = interp1(Vtn_prime,squeeze(y1),Vtn);
Var_Interp.([Name_Cond{k}]).Cinem.(Foot{k}{f}).(CinemVar_Names.([Name_Cond{k}]).(Foot{k}{f}){v}){o,1} = interp1(Vtn3,squeeze(y3),Vtn);
MeanPressure.([Name_Cond{k}]).(Foot{k}{f})(j,i)=nanmean(MeanPressure_Concatenate.([Name_Cond{k}]).Pressions_Bis.(Foot{k}{f})(j,i,:));
StdPressure.([Name_Cond{k}]).(Foot{k}{f})(j,i)=nanstd(MeanPressure_Concatenate.([Name_Cond{k}]).Pressions_Bis.(Foot{k}{f})(j,i,:));

Source: Script_Dynamisme.m — 2017

CellStat=[Cell0,Cell,Cell1];
CellName=[Cell0_Name,Cell_RatioNameSignal,Cell_NameSignal];
save(fullfile(Direct_VarSaved,'Save_Stat_Table.mat'),'CellName','CellStat','-v7.3')
filename=fullfile('<REDACTED>','Dyna_Anova.xls');
xlwrite(filename, CellName(1:20), 'Tableau Anova','A1')
xlwrite(filename, [CellStat{1:20}], 'Tableau Anova','A2')

• Oxelo: the tabulated outputs show a systematic exploration of head-tube angles, handlebar heights, and joint-constraint combinations; the framing material specifies the variables to be measured, the tasks, and the equipment. This work demonstrates the ability to turn a product question into tunable parameters, biomechanical criteria, and a testable protocol — a core early-stage R&D skill. Source: Angle chasse_Obj90.txt — 2016/2017

Best (deg)   Angle chasse (deg)   Angle Pied3 (deg)   Angle Genou3 (deg)
90.683448343 90.683448343         80                  5
89.932290568 89.932290568         80.042858817        11.578329351

• Quechua: the R scripts implement a clear statistical workflow for the mechanical test. Dataset import, reference-shoe definition, linear modeling with lm, global ANOVA, pairwise comparisons, export of p-value matrices, then multivariate analysis through PCA followed by HCPC to group shoes by mechanical behavior are all represented. This is a strong example of moving from a test bench to a statistically justified product ranking through inferential statistics and unsupervised classification. Source: Anova_Quechua.R — 2016/2017

Ext_ANOVA <- read.delim2("<REDACTED>/Table_Anova2.txt", dec=".")
Ext_ANOVA$Echantillons <- relevel(Ext_ANOVA$Echantillons, NaneEchantillon[i])
mod=lm(Vitesse~ Echantillons , data=Ext_ANOVA)
Anova(mod, type="II")
summary(mod)
p=pairwise.t.test(Ext_ANOVA$Amplitude, Ext_ANOVA$Echantillons, p.adj="none")
write.table(p$p.value, "Pairwise_Comparison_Amplitude2_final.Xls" , sep="\t", dec=".")

Source: Classification_TestMeca.R — 2016/2017

Test.Meca <- read.delim2("<REDACTED>/Test Meca.txt",row.names = 1)
res.pca=PCA(Test.Meca)
res.hcpc=HCPC(res.pca,nb.clust=5)
Test.Meca2=Test.Meca2[,c(1,2)]
res.pca2=PCA(Test.Meca2)
res.hcpc2=HCPC(res.pca2,nb.clust=5,consol=TRUE,graph=TRUE)
plot(res.hcpc2, choice="tree", angle=60)

🧠 Interpretation and scientific communication

• Domyos: the scripts and presentation translate the signals directly into product messages, for example comparing the effects of Kangaroo and Twist modes on energy expenditure and on specific muscle groups. This shows the ability to move from EMG/physiology processing to a decision-oriented concept discussion.
• Kalenji: the visible work turns a literature review and a Footscan reliability audit into field-use recommendations. That is valuable scientific communication because it connects percentages, cross-study comparisons, and measurement biases to practical in-store guidance.
• NewFeel: the posters, compiled videos, and pressure maps clearly aim to make complex phenomena readable by non-developers. The work highlights the ability to make continuous signals and pressure matrices understandable in terms of comfort, load distribution, ideal thresholds, and future product uses.
• Oxelo: the framing material reformulates the topic into tasks, observed variables, experiment time, instrumentation, and budget. This shows the ability to make scientific intent actionable for project teams.
• Quechua: the combination of mechanical test, ranking, and shortlist turns statistics and clusters into a clear product decision. This is exactly the kind of translation expected by a product team that does not want to read the R code.

🧾 Deliverables

• Domyos: Matlab scripts for Fitmate/EMG processing, comparison figures, Excel exports, and an internal presentation on the prototype's energy and muscle effects.
• Kalenji: structured literature review (~40 articles) on foot strike patterns covering comparative biomechanics of RFS/MFS/FFS, injuries, running economy, and shoe-property influence; in-store Footscan dataset audit (28,200 scans) and reliability/protocol recommendations.
• NewFeel: Matlab scripts for multimodal analysis, VarSaved datasets, ANOVA exports, pressure maps, posters, video compilations, and synthesis decks on ideal cushioning and foot roll-over.
• Oxelo: V1/V2 protocols, budget estimate, list of biomechanical variables, parametric optimization outputs, and study-framing material.
• Quechua: R scripts for ANOVA and classification, pairwise comparison tables, shoe ranking, mechanical-results presentation, and a biomechanical-study proposal.

📈 Results

• Based on the dated artifacts provided, this work structured five distinct projects ranging from physiological analysis and signal processing to statistical classification and experimental framing. It resulted in reusable scripts, analysis tables, visualizations, presentations, and protocols that could be used directly by the brands. The overall setup forms a coherent R&D chain: acquisition, cleaning, comparison, ranking, and business-facing restitution. Benefit: reducing the time between raw measurement and product decision while making scientific interpretation more robust.

Project Domyos

• Period: July 2016 to November 2016, based on a Matlab auto-generation date and the presentation dated November 10, 2016.
• Deliverables / outputs: Matlab scripts for Fitmate and EMG processing, comparisons across Classic, Kangaroo, and Twist modes, Excel exports, and an internal oral presentation. Benefit: quickly objectifying the prototype's energy and muscle trade-offs without relying on manual post-processing.

Project Kalenji

• Period: 2016.
• Deliverables / outputs: literature review (~40 articles) on FSP, synthesis of the 5 main distribution studies (cohorts of 248 to 1,991 runners), in-store Footscan dataset analysis (28,200 scans), and reliability/protocol recommendations. Benefit: securing the use of an in-store tool by making its limits and minimum operating conditions explicit.

Project NewFeel

• Period: April 2017, with 2017 artifacts covering both the ideal-cushioning study and the fundamental foot roll-over study.
• Deliverables / outputs: Matlab scripts for multimodal analysis, ANOVA tables, pressure maps, posters, video compilations, and synthesis decks on sensory ideal and foot mechanics. Benefit: connecting lab measurements, user perception, and biomechanical interpretation in a format directly useful for footwear innovation.

Project Oxelo

• Period: December 2016 to February 2017, based on the framing material and meeting notes present in the folder.
• Deliverables / outputs: study protocols, observed variables, budget, task criteria, and parametric optimization outputs around product geometry. Benefit: de-risking a future study by establishing a measurable methodological framework before final protocol execution.

Project Quechua

• Period: September 2016 to May 2017, based on the initial framing artifacts and Visual3D-related pipelines dated 2017.
• Deliverables / outputs: inversion mechanical test, statistical and classification R scripts, shortlist of shoes for the next biomechanical phase, and reporting material. Benefit: turning a mechanical bench test into an evidence-based product ranking and a rational shortlist for deeper study.

🔧 Technical environment

• Matlab is dominant on Domyos and NewFeel for multi-file import, filtering, interpolation, multi-sensor synchronization, figure generation, video generation, and Excel export; R is used on Quechua for ANOVA, post-hoc comparisons, PCA, and HCPC clustering. Instrumentation and ecosystem visible in the artifacts include Delsys EMG, Fitmate Pro/Cosmed, Footscan/RSscan, Qualisys, Kistler, Visual3D/V3D, spreadsheets, and PDF/PPT deliverables for internal diffusion.