Riba:
- Ability don gina dabaru na fusion (mai haɗawa / Kalman tace) wanda ya haɗu da IMU, encoder da bayanan firikwensin nesa tare da AI
- Ikon daidaitawa da fassara rafukan bayanan IoT na tushen MQTT da telemetry tare da taimakon AI
- Ƙarfin ganowa da daidaita sautin firikwensin, daidaitawa da al'amurran daidaita lokaci a cikin fitarwa na AI
Babu firikwensin da yake cikakke. Gyroscope yayi daidai sosai a cikin ɗan gajeren lokaci, amma yana shuɗewa akan lokaci. Accelerometer ya tsaya tsayin daka amma yana samun hayaniya daga girgiza. Encoder yana auna matsayi daidai, amma ba daidai ba ne idan akwai zamewa. GPS yana aiki a cikin faffadan yanki amma yana da ƙarancin mitoci da hayaniya. Injiniyan mechatronic dole ne ya samar da gaskiya guda ɗaya, abin dogaro daga waɗannan na'urori marasa kamala; Wannan shine haɗakar firikwensin. Bugu da ƙari, a cikin tsarin zamani, wannan bayanan ba kawai ya kasance na gida ba, amma yana gudana zuwa gajimare da cibiyoyin sarrafawa ta hanyar cibiyoyin sadarwa na IoT. Hankali na wucin gadi taimako ne mai ƙarfi a cikin duka saita algorithms fusion da daidaitawa da fassarar telemetry na IoT. A cikin wannan rukunin, mun rufe matattara mai dacewa, Kalman tace dabaru da yadda ake saita kwararar IoT na tushen MQTT tare da AI da kama hayaniya da tarko na aiki tare.
Me yasa Fusion Sensor?
Kowane firikwensin yana da "yankin mitar mai kyau". Manufar haɗakarwa ita ce yin amfani da kowane firikwensin a yankin da yake da ƙarfi da kuma tallafa masa tare da ɗayan a yankin da yake da rauni.
firikwensin
karfi batu
rauni
gyroscope
Gudun kusurwa na gajeren lokaci, amsa mai sauri
Juyawa akan lokaci
accelerometer
Maganar gangare na dogon lokaci (nauyi)
Jijjiga / hayaniya akan motsi
encoder
Wuri mai tsayi
Kuskuren zamewa/ koma baya
GPS
cikakken matsayi
Ƙananan mita, amo, babu na cikin gida
Dogaro da firikwensin guda ɗaya yana gabatar da raunin wannan firikwensin a cikin tsarin. Fusion, alal misali, yana haɗa kusurwar gyroscope mai sauri amma mai jujjuyawa tare da jinkirin amma tsayayyen tunani na ma'aunin accelerometer, yana samar da kusurwa mai sauri kuma mara faɗuwa.
Tace mai kari
Mafi sauƙaƙa kuma mafi yawan gama gari shine matattarar ƙarin. Tunanin yana da hankali: amince da gyroscope a babban mitar (sauyi mai sauri), accelerometer a ƙananan mitar (jinkirin, barga). Takaitacciyar layi ɗaya:
# Karin tacewa: kimanta jin zafi (pitch) # alpha ~ 0.98: ƙari ga gyroscope, ƙasa da ma'aunin accelerometer (digiri/ sec) accelerometer: kusurwar da aka lasafta daga accelerometer (digiri) dt : lokacin samfur (sec) """ gyro_gytgytscope = baya + gyro_gytscope + * gyro_pace + (1 - alpha) * acceleration_angle# Tabbatar da hankali: tsantsar gyroscope (drifts) idan alpha = 1, # idan alpha = 0 tsarkin accelerometer (mai hayaniya).
Ƙididdigar alpha a nan yana ƙayyade ma'auni: idan yana kusa da 1, yana dogara ne akan gyroscope (hadari na haɓakawa), idan yana kusa da 0, ya dogara da accelerometer (ƙarar ƙararrawa). Matsakaicin ƙima shine 0.95-0.98. AI na iya ba da shawarar wannan ƙimar, amma madaidaicin alpha ya dogara da ƙimar samfurin tsarin ku da halayen amo; an saita shi azaman gwaji.
Tukwici: Lokacin buga matattara ga AI, tabbatar da tambayar yadda ake auna dt. Yawancin kurakurai suna zuwa ne daga ɗauka cewa dt yana dawwama amma a zahiri lokacin zagayowar ya bambanta. Auna dt na gaske tare da millis()/timestamp, kar a rubuta madaidaicin.
Kalman Tace Logic
Madaidaicin tace mai sauƙi ne amma baya ƙididdige ƙididdiga na amo. Tacewar Kalman tana samar da ingantacciyar ƙima (ƙarƙashin wasu zato) ta hanyar ƙirƙira hayaniyar firikwensin ƙila da aiwatar da rashin tabbas. Yana aiki a matakai biyu:
- Hasashen: Yi hasashen yanayi na gaba da rashin tabbas tare da tsarin tsarin.
- Sabuntawa: Auna sabon ma'auni akan amincin ma'auni (ribar Kalman) kuma gyara kimanta.
Kalman gain K yana amsa tambayar ta atomatik "Shin in amince da ma'auni ko samfurin" a kowane mataki. Idan amo yana da girma, K ya zama ƙarami (amincewa a cikin samfurin), idan rashin tabbas na tsari ya yi girma, K ya zama mafi girma (kwarin gwiwa a cikin ma'auni). AI cikin sauƙi yana rubuta matattar Kalman mai girma ɗaya; amma aikin ku ne ku zaɓi ƙungiyoyin amo (`Q`, `R`) daidai kuma sun fito ne daga ainihin hayaniyar tsarin.
Hankali: Q (hayaniyar tsari) da R (ƙaramar aunawa) ƙimar da AI ta bayar sune samfurin / mai ɗaukar wuri. Idan ba ku ƙayyade waɗannan daga ainihin amo na tsarin ba (misali bambance-bambancen da aka auna lokacin da firikwensin ya tsaya), tacewa zai kasance ko dai a hankali ko kuma amo. Kada ku amince da haɗin kai da AI ta bayar a matsayin "ƙimar gaskiya".
IoT Data Rafi: MQTT da Telemetry
Tsaftataccen bayanan da kuke samarwa tare da fusion yawanci ana jigilar su zuwa cibiyar ta hanyar hanyar sadarwa ta IoT. Ka'idar da aka fi sani da ita a cikin masana'antar ita ce MQTT: nauyi mai nauyi, samfurin buga-biyan kuɗi, dace da ƙananan bandwidth. AI da sauri ta kafa MQTT mai wallafa/lambar mai biyan kuɗi da tsarin tsarin telemetry na JSON.
shigo da json, timeimport paho.mqtt.client as mqttclient = mqtt.Client () abokin ciniki.connect ("broker.local", 1883, keepalive=60) def telemetry_broadcast (zazzabi, zazzabi, vibration_rms): saƙo = {"ts": time.time(), # timestampronpa "condiction" -" 2), "temperature_c": zagaye (zazzabi, 1), "vibration_rms": zagaye (temperature_rms, 3), "unit": {"pain": "deg", "temperature": "C", "vibration": "mm/s"} } client.publish("machine/jline1/dusor), qos=1: isar da aƙalla sau ɗaya. Yin amfani da qos=0 (zai iya ɓacewa) akan mahimman bayanai.
Hukunce-hukuncen injiniya guda biyu suna da mahimmanci a nan: (1) tambarin lokaci (ts) kowane saƙo - ana ƙididdige firikwensin firikwensin daban-daban a lokuta daban-daban, aiki tare yana yiwuwa ne kawai tare da tambarin lokaci; (2) Matakin QoS-mahimman bayanai yana amfani da aƙalla qos=1 (an isar da shi aƙalla sau ɗaya), qos=0 yana ba da damar asarar saƙo.
Surutu, Daidaitawa da Aiki tare
Komai kyawun yadda haɗin ke da kyau, matsaloli uku tare da shigarwar suna karkatar da sakamakon:
- Surutu: Idan danyen bayanan firikwensin ya shiga cikin fusion ba tare da an tace shi ba, fitarwar fusion shima zai yi hayaniya. Ana iya buƙatar kafin tacewa (matsakaici, ƙaramin wucewa).
- Daidaitawa: Kuskuren kashewa da ma'auni (misali sifilin sifilin na'urar accelerometer) yana ɓatar da haɗin kai tsaye. Dole ne a daidaita na'urori masu auna firikwensin kafin amfani.
- Aiki tare na lokaci: Ana buƙatar ɗaukar lokaci da haɗin gwiwa don daidaita na'urori masu auna firikwensin da aka ƙididdige su a farashi daban-daban (misali GPS 10 Hz tare da 1 kHz IMU).
AI na iya ƙara waɗannan matakan zuwa lambar, amma ko ana buƙatar kowannensu kuma sigoginsa sun keɓance ga tsarin ku. A cikin fitowar AI "ina ne calibration?" kuma "suna daidaita lokutan lokutan?" Tabbatar yin tambayoyi.
Rauni Mai Sauƙi / Ƙarfi Mai ƙarfi
RAUNI: "Kirga kwana daga bayanan IMU." (Wanne tacewa? Ana sarrafa drift? Ta yaya dt? Ba a bayyana ba.) KARFI: "Haɗa bayanan accelerometer da gyroscope daga MPU6050 tare da tacewa da ƙididdige kusurwar filin wasa. Yi ƙididdige dt daga ainihin timestamp, kar a ɗauka akai-akai. Saita alphalt tare da ma'auni 9). Tane2
Mini Case
Injiniyan Robotics Selin yana son matatar da ta dace daga AI don kimanta kusurwa akan na'urar daidaita ƙafafu biyu. AI yana ba da lamba mai tsabta, amma robot a hankali yana jingina gefe ɗaya. Selin yana duba dt: lambar tana ɗauka dt ɗin koyaushe 0.01 s ne, yayin da lokacin zagayowar ya bambanta saboda na'urorin Bluetooth. Lokacin da aka ƙididdige dt daga tambarin lokaci na ainihi, ɗigon yana raguwa amma ba ya ɓace gaba ɗaya. Sa'an nan ya dubi calibration na accelerometer; Ko da lokacin da firikwensin ya tsaya ana samun 2° diyya. Lokacin da na cire diyya na daidaitawa, robot ɗin yana tsaye tsaye. A ƙarshe, yana lura cewa yana amfani da qos = 0 lokacin watsa shirye-shiryen telemetry ta MQTT kuma yana ƙara shi zuwa qos = 1 don mahimman bayanan kusurwa. AI ya isar da kwarangwal a cikin mintuna; amma tabbacin injiniyan ya kama wasu takamaiman batutuwa guda uku: m dt, daidaitawa, da QoS.
Kuskuren gama gari
- Zaton dt yana dawwama, yayin da lokacin sake zagayowar ya bambanta (fusion drifts).
- Fusing na'urori masu auna sigina ba tare da calibrating su (tsari biya diyya).
- Ƙoƙarin daidaita na'urori masu auna gudu daban-daban ba tare da buga lokaci ba.
- Barin tare da ƙimar samfurin AI ba tare da aunawa Kalman Q / R covariances daga ainihin amo.
- Ba da izinin asarar saƙo ta amfani da qos=0 akan mahimman bayanan IoT.
- Bayar da ɗanyen bayanan hayaniya zuwa gaɓoɓin ba tare da an riga an tacewa ba.
A takaice
- Fusion yana amfani da kowane firikwensin a cikin yankin mitar inda yake da ƙarfi kuma yana rama rauninsa.
- Tace mai dacewa yana da sauƙi; Yana kafa ma'auni na gyroscope-accelerometer tare da alpha.
- Kalman tace samfurin hayaniya mai yiwuwa; Q/R ya fito daga ainihin tsarin.
- A cikin telemetry na MQTT, tambarin lokaci da QoS masu dacewa sune yanke shawara mai mahimmancin injiniya.
- Hayaniya, daidaitawa da aiki tare da lokaci suna ƙayyade ingancin haɗuwa.
- dt ana auna shi daga ainihin lokaci; an tabbatar da calibration da QoS ba tare da barin shi ga AI ba.
Aikin aikace-aikace
Don IMU (bayanan gyro na ainihi ko simulated + accelerometer) suna da AI suna samar da matatar haɗin kai kuma tabbatar cewa an ƙididdige dt daga ainihin tambarin lokaci. Sa'an nan: (1) gwada alpha don 0.90, 0.98 da 1.0 kuma kula da drift da amo balance, (2) ƙara ƙaddamarwa da gangan (kuskuren daidaitawa) zuwa firikwensin kuma duba yadda fitarwar fusion ɗin ke motsawa, (3) canza bayanan zuwa tsarin MQTT JSON kuma ƙara timestamp da filayen naúrar. Yi la'akari da wane ƙimar alpha ne ke ba da mafi daidaiton sakamako don bayanan ku da nawa daidaitawar daidaitawa ke karkatar da fitarwa.