Realtime Predictive Maintenance with Lambda Architecture

ࣾஂ๏ਓ ిࢠ৘ใ௨৴ֶձ THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS ৴ֶٕใ TECHNICAL REPORT OF IEICE. SC2016-28 (2016-11) ϥϜμΞʔΩςΫνϟʹΑΔϦΞϧλΠϜ༧ଌอक ࢁొ༱࣍† ۽࡚޺थ† ෱ຊՂ࢙† † NTT ιϑτ΢ΣΞΠϊϕʔγϣϯηϯλ ౦ژ౎෢ଂ໺ࢢ྘ொ3-9-11 E-mail: †{yamato.yoji,kumazaki.hiroki,fukumoto.yoshifumi}@lab.ntt.co.jp ͋Β·͠ ۙ೥ɼIoT ٕज़͕ਐల͓ͯ͠Γɼϝϯςφϯε෼໺ͰͷԠ༻͕ظ଴͞Ε͍ͯΔɽ͔͠͠ɼݱ৔ঢ়گΛϦΞ ϧλΠϜʹ෼ੳͰ͖͍ͯͳ͍ɼσʔλऩूͷίετ͕ߴ͍ɼނোݕ஌ϧʔϧͷઃఆίετ͕ߴ͍౳ͷ՝୊͔Βɼ೔ຊ Ͱ͸े෼޿͕͍ͬͯͳ͍ɽຊߘ͸ɼ͜ΕΒ՝୊ղܾͷͨΊɼϥϜμΞʔΩςΫνϟͷߟ͑Λ༻͍ͯɼΤοδ͸ηϯα σʔλΛղੳ͠ϦΞϧλΠϜʹΞϊϚϦʔΛݕ஌͠ɼ৽ͨͳϧʔϧΛநग़͢Δͱͱ΋ʹɼΫϥ΢υ͸όονͰऩू͞ ΕͨσʔλΛ෼ੳ͠ΤοδͷֶशϞσϧΛΑΓߴ͍ਫ਼౓ʹߋ৽͢ΔɼϝϯςφϯεϓϥοτϑΥʔϜΛఏҊ͢Δɼߋ ʹɼαϯϓϧΞϓϦέʔγϣϯΛ࣮૷͢Δɽ Ωʔϫʔυ Jubatusɼ༧ଌอकɼIoTɼϥϜμΞʔΩςΫνϟɼΫϥ΢υίϯϐϡʔςΟϯάɼIndustrie 4.0ɼ Realtime Predictive Maintenance with Lambda Architecture Yoji YAMATO†, Hiroki KUMAZAKI†, and Yoshifumi FUKUMOTO† † Software Innovation Center, NTT Corporation 3-9-11, Midori-cho, Musashino-shi, Tokyo 1808585 Japan E-mail: †{yamato.yoji,kumazaki.hiroki,fukumoto.yoshifumi}@lab.ntt.co.jp Abstract Recently, IoT technologies have been progressed and applications of maintenance area are expected. However, IoT maintenance applications are not spread in Japan yet because of insufficient analysis of real time situation, high cost to collect sensing data and to configure failure detection rules. In this paper, using lambda architecture concept, we propose a maintenance platform in which edge nodes analyze sensing data, detect anomaly, extract a new detection rule in real time and a cloud orders maintenance automatically, also analyzes whole data collected by batch process in detail, updates learning model of edge nodes to improve analysis accuracy. Key words Jubatus, Predictive Maintenance, IoT, Lambda Architecture, Cloud Computing, Industrie 4.0, 1. ͸ ͡ Ί ʹ ۙ೥ɼIoTʢInternet of Thingsʣٕज़΍Ϋϥ΢υٕज़ʢྫ͑ ͹ [1]ʣ͕ਐల͍ͯ͠ΔɽIoT ͷద༻ൣғ͸ଟذʹ౉Δ͕ɼͦͷ தͰ΋ɼυΠπͷ Industrie 4.0 [3] ߏ૝ͰਐΊΒΕ͍ͯΔɼ੡଄ ΍ϝϯςφϯε෼໺͕ద༻ઌͱͯ͠༗ྗࢹ͞Ε͍ͯΔɽ޻৔ɼ ػցɼ੡඼౳ͷ৘ใΛηϯαʹΑΓऩूɼ෼ੳ͢Δ͜ͱͰɼͦ ΕΒͷঢ়ଶΛՄࢹԽ͠ɼੜ࢈೺Ѳɼܭը൓өɼ෺ྲ੍ྀޚɼෆྑ ෺඼ަ׵౳ͷαϓϥΠνΣʔϯͷࣗಈԽɼαʔϏε࿈ܞʢྫ͑ ͹ [2]ʣʹΑΔɼϏδωεͷՃ଎͕ظ଴͞Ε͍ͯΔɽ IoT σʔλΛ༻͍ͨɼIoT ΞϓϦέʔγϣϯΛ։ൃɼӡ༻͢ ΔͨΊͷɼIoT ϓϥοτϑΥʔϜ΋ग़͖͍ͯͯΔ [4], [5]ɽ͔͠ ͠ɼطଘͷ IoT ϓϥοτϑΥʔϜ͸ɼେྔͷηϯασʔλΛऩ ू͠ՄࢹԽ͢Δ͜ͱ͕ओͳ஫ྗ఺ͱͳ͓ͬͯΓɼલड़ͷΑ͏ͳ ϝϯςφϯεͷՃ଎ʹੜ͔͢͜ͱ͸े෼͞Ε͓ͯΒͣɼ೔ຊͰ ͸े෼ීٴ͍ͯ͠ͳ͍ɽ۩ମతʹ͸ɼݱ৔ঢ়گΛϦΞϧλΠϜ ʹ෼ੳͰ͖͍ͯͳ͍ɼσʔλऩूͷωοτϫʔΫίετ͕ߴ͍ɼ ނোݕ஌ϧʔϧͷઃఆίετ͕ߴ͍໰୊্͕͛ΒΕΔɽ ຊߘͰ͸ɼ޻৔Ͱͷۀ຿ػثΛ୊ࡐʹɼ͜ΕΒ՝୊Λղܾ͢ ΔϝϯςφϯεϓϥοτϑΥʔϜΛఏҊ͢ΔɽఏҊϓϥοτ ϑΥʔϜ͸ɼΤοδ͸ηϯασʔλΛղੳ͠ϦΞϧλΠϜʹΞ ϊϚϦʔΛݕ஌͠৽ͨͳϧʔϧΛநग़͢Δͱͱ΋ʹɼΫϥ΢υ ͸όονͰऩू͞ΕͨσʔλΛ෼ੳ͠ΤοδͷֶशϞσϧΛΑ Γߴ͍ਫ਼౓ʹߋ৽͢ΔɽߋʹɼఏҊϓϥοτϑΥʔϜ্ʹɼα ϯϓϧΞϓϦέʔγϣϯΛ࣮૷͢Δɽ 2. طଘٕज़ͷ՝୊ طଘIoT ϓϥοτϑΥʔϜٕज़ͷ՝୊Λ੔ཧ͢Δɽ AWS IoT [4] ͸ɼAmazon Web Services ͷ֤छػೳΛ IoT Ͱ ౷߹ར༻Մೳʹ͢ΔϓϥοτϑΥʔϜͰ͋Δɽྫ͑͹ɼAmazon — 49 — Kinesis Λ༻͍ͯɼMQTT(MQ Telemetry Transport) ϓϩτ ίϧͰσʔλΛऩू͠ɼऔूͨ͠σʔλΛ Amazon Machine Learning ͷػցֶशػೳʹΑΓճؼ΍Ϋϥε෼ྨͷଟ࠼ͳ෼ ੳ౳͕ՄೳͰ͋Δɽ NTT υίϞͱ GEʢGeneral Electricʣࣾ͸ɼGE ͷ࢈ۀ༻ ػث޲͚ϫΠϠϨεϧʔλʔ Orbit ͱɼυίϞͷ௨৴Ϟδϡʔ ϧΛ࿈ܞͤͨ͞ IoT ιϦϡʔγϣϯΛ 2015 ೥ʹൃද͍ͯ͠ Δ [5]ɽاۀ͸ɼԕִ஍ͷઃඋʹυίϞͷ௨৴ϞδϡʔϧΛ಺ ଂͨ͠ GE ͷ Orbit Λઃஔ͢Δ͜ͱͰɼઃඋͷՔಇσʔλΛऩ ूͰ͖Δɽ͞ΒʹɼυίϞ͕ఏڙ͢Δ Toami ্Ͱ IoT ΞϓϦ έʔγϣϯΛ։ൃՄೳͰ͋ΓɼऔಘσʔλͷՄࢹԽ΍ Web αʔ Ϗε࿈ܞʢ [6] [7] ౳ʣΛ༰қʹ͍ͯ͠Δɽ ͔͠͠ɼ͜ΕΒͷٕज़Λϝϯςφϯεʹར༻͢Δࡍ͸ɼେ͖ ͘ 3 ͭ՝୊͕͋Δͱߟ͑Δɽ ୈҰʹݱ৔ঢ়گΛϦΞϧλΠϜʹ෼ੳͰ͖͍ͯͳ͍ɽ[5] ͸ɼ ू໿σʔλͷόονॲཧʹΑΔՄࢹԽ͕த৺Ͱ͋ΓɼϦΞϧλΠ Ϝͳ෼ੳʹجͮ͘෦඼ख഑౳ʹ͸ܨ͕͍ͬͯͳ͍ɽIndustrie4.0 Ͱ͸ɼࣄલʹఆٛͨ͠ҟৗϧʔϧʹجͮ͘ྲྀ௨ख഑౳͸૝ఆ͞ Ε͍ͯΔ͕ɼ޻৔ͷ؀ڥ΍قઅ౳ͷ৚݅Ͱద੾ͳϧʔϧ͕มΘ Δݱ৔ঢ়گʹ௥ै͢Δ͜ͱ͸Ͱ͖͍ͯͳ͍ɽ ୈೋʹηϯγϯάσʔλऩूͷίετ͕ߴ͍ɽAWS IoT Ͱ ͸෼ੳͷͨΊҰ౓શσʔλΛΫϥ΢υʹूΊΔ͕ɼଟ͘ͷػց ͷηϯασʔλΛɼ֤஍͔Βऩू͢ΔͨΊͷωοτϫʔΫ͕ඞ ཁͰ͋Δɽྫ͑͹ɼIoT ʹΑΔϝϯςφϯεࣄྫͱͯ͠ஶ໊ͳ ίϚπࣾͷ KOMTRAX ͸ɼݐઃػցͷσʔλऩूʹӴ੕௨৴ Λ࢖͓ͬͯΓɼίετ͕େ͖͍ͱڞʹɼશͯͷσʔλ͸ૹΕͣ σʔλΛؒҾ͍ͯૹ͍ͬͯΔɽ ୈࡾʹނোݕ஌ͷͨΊͷϧʔϧઃఆͷίετ͕ߴ͍ɽ༷ʑͳ IoT ηϯασʔλ͔ΒɼނোΛݕग़͢ΔͨΊʹ͸ɼ෼ੳΞϓϦ έʔγϣϯଆͰϧʔϧ΍ᮢ஋Λઃఆ͠൑ఆ͢Δ͜ͱ͕ओྲྀͰ͋ Δɽྫ͑͹ɼPSPP [8] ౳ͷ౷ܭղੳιϑτ΢ΣΞΛ༻͍ͯɼͦ ͷ൑ఆϧʔϧΛநग़͢Δํ๏͕͋Δ͕ɼϧʔϧɼᮢ஋ઃఆ͸ߴ ౓ͳ஌͕ࣝඞཁͰ͋Γɼͦͷઃఆίετ͸ߴ͍ɽ ͜ΕΒͷ՝୊͔Βɼ೔ຊͰ͸ϝϯςφϯε෼໺ʹ IoT ͕े෼ ීٴ͍ͯ͠Δͱ͸ݴ͑ͳ͍ɽ 3. ϥϜμΞʔΩςΫνϟίϯηϓτʹجͮ͘ϝ ϯςφϯεϓϥοτϑΥʔϜͷఏҊ લઅͷ՝୊Λղܾ͢ΔͨΊɼlambda ΞʔΩςΫνϟͷίϯη ϓτΛ༻͍ͨϝϯςφϯεϓϥοτϑΥʔϜΛఏҊ͢ΔɽϥϜ μΞʔΩςΫνϟ͸ɼ෼ੳ݁ՌΛϢʔβʹఏڙ͢Δࡍʹɼόο νϨΠϠʔɼεϐʔυϨΠϠʔͷ྆ํͰ෼ੳ͢Δ͜ͱͰɼࡉ΍ ͔ͳूܭ݁Ռͱ଎ใੑͷඞཁͳ݁ՌΛ࣮ݱ͢ΔɼMarz ͕ఏҊ ͨ͠ΞʔΩςΫνϟͰ͋Δɽ[9] ਤ1 ʹϥϜμΞʔΩςΫνϟͷద༻ΠϝʔδΛࣔ͢ɽ޻৔౳ ͷΤοδଆʹ͸ɼσʔλอ࣋༻ͷετϨʔδʹՃ͑ɼJubatus [10] ͕഑ஔ͞ΕΔɽJubatus ͸ɼετϦʔϜσʔλͷஞ࣍ॲཧʹద ͨ͠ػցֶशϑϨʔϜϫʔΫͰ͋ΔɽΫϥ΢υଆʹ͸ɼCEP ʢComplex Event Prosessing), DB, ϝϯςφϯεΞϓϦέʔ γϣϯ͕഑ஔ͞ΕΔɽ ·ͣɼεϐʔυϨΠϠʔͰ͸ɼΤοδଆͷ Jubatus ͕ɼηϯ ασʔλΛετϦʔϜॲཧ͠ɼҟৗ͍ٙͷ͋ΔσʔλΛݕ஌͠ɼ ͦͷ৔߹ʹΫϥ΢υଆͷ CEP ʹσʔλΛૹ৴͢ΔɽJubatus ͸ػցֶशΛ௨ͯ͡৽ͨͳҟৗ͍ٙϧʔϧ΋நग़ͯ͠ૹ৴͢Δɽ CEP Λհͯ͠ҟৗ͍ٙσʔλΛड͚औͬͨϝϯςφϯεΞϓ Ϧέʔγϣϯ͸ɼނো༧ଌ౳ͷ෼ੳΛ͠ɼඞཁʹԠͯ͡֎෦γ εςϜʢERPʢEnterprise Resource Planningʣ౳ʣͱ࿈ܞ͢ Δɽ֎෦γεςϜ࿈ܞ͸ɼWeb αʔϏε౳ͷطଘ࿈ܞٕज़ʢྫ ͑͹ [11]ʣΛ༻͍Ε͹Α͍ɽ࣍ʹɼόονϨΠϠʔͰ͸ɼΤο δଆͷੜσʔλ͕ɼ໷ؒ౳ͷίετ͕͍҆࣌ؒʹɼΫϥ΢υଆ ʹૹ৴͞ΕɼDB ʹอ࣋͞ΕΔɽϝϯςφϯεΞϓϦέʔγϣ ϯ͸ɼεϐʔυϨΠϠʔͰ͸ର৅֎ͷৄࡉͳ෼ੳΛɼશσʔλ Λ༻͍ͯ෼ੳ͢ΔɽߋʹɼσʔλαΠΤϯςΟετ͸ɼఆظత ʹɼੜσʔλΛ෼ੳͯ͠ɼΑΓਫ਼౓ߴ͍ػցֶशϞσϧΛநग़ ͠ɼΤοδଆʹ഑৴͢Δ͜ͱ΋Ͱ͖ΔɽΤοδଆͷ෼ੳਫ਼౓͕ े෼ߴ͘ͳͬͨࡍʹ͸ɼόονϨΠϠʔ͸ෆཁͱͯ͠΋Α͍ɽ ਤ1 ͷΞΠσΞʹΑΓɼલઅ՝୊͕ղܾ͞ΕΔɽJubatus ʹ ΑΓϦΞϧλΠϜͰҟৗ͍ٙΛݕ஌͠ɼަ׵෦඼ൃ஫౳ͷଈ ࣌ͷΞΫγϣϯʹܨ͛Δ͜ͱ͕Ͱ͖Δɽ·ͨɼJubatus ͸ɼ௨ ৗͷσʔλ஋ͱҟͳΔ৔߹Λݕ஌͢ΔΞϊϚϦʔݕ஌ʢLOF ʢLocal Outlier Factorʣ[12] ౳ʣ౳ͷΞϧΰϦζϜ͕ར༻Ͱ͖ ΔͨΊɼݱ৔؀ڥσʔλʹԠͨ͡ϧʔϧ΍ᮢ஋Λநग़͠ɼΫϥ ΢υʹૹ৴͠ɼϝϯςφϯεΞϓϦέʔγϣϯʹ൓өͰ͖Δɽ εϐʔυϨΠϠʔͰ͸ɼҟৗ͍ٙσʔλͷΈૹ৴͢Δ͜ͱͰ ωοτϫʔΫίετΛԼ͛Δ͜ͱ͕Ͱ͖ΔɽόονϨΠϠʔͰ ඞཁͳશσʔλ͸ɼ໷ؒ΍෺ཧత༌ૹ౳ͷखஈͰɼίετΛԼ ͛Δ͜ͱ͕Ͱ͖Δɽ ਤ2 ͸ɼ্هΞΠσΞʹج͍ͮͨఏҊϓϥοτϑΥʔϜͷ ΞʔΩςΫνϟͰ͋Δɽਤ2 Λࢀরͯ͠ɼॲཧεςοϓΛઆ໌ ͢Δɽ 1. Τοδʹ͸ɼۀ຿ػثʹηϯαʔ͕ઃஔ͞Εɼηϯγϯ άσʔλ͕ਵ࣌ऩू͞Εอଘ͞ΕΔɽΤοδϊʔυͷ Jubatus ͸ɼετϦʔϜॲཧͰσʔλΛΞϊϚϦʔݕ஌΍෼ྨ෼ੳΛ͠ɼ ௨ৗӡ༻ͱେ͖͘ҟͳΔ৔߹ʹɼҟৗΛ൑ఆ͠ɼΫϥ΢υଆʹ ؔ࿈ͨ͠σʔλΛૹ৴͢ΔɽJubatus ͸ΦϯϥΠϯͰ൑ఆ͚ͩ Ͱͳ͘ɼֶश΋Ͱ͖ΔͨΊɼ৽ͨͳҟৗ͍ٙϧʔϧΛநग़ͨ͠ ৔߹͸ɼͦΕ

…(Full text truncated)…

This content is AI-processed based on ArXiv data.