PowerPoint Presentation - PDF Free Download

AGENDA 1 2 3 4 5 분석에대한니즈의변화 SAP Analytics 요약 SAP PA의자동화방법론및적용데모 사례결언

OLAP Ad-hoc Query EDW Predictive Modeling 최적화

CPU : 1,920 개 / GPU : 280 개 Monte Carlo Tree Search CNN(Convolution Neural Network) 참조 : https://www.tastehit.com/blog/google-deepmind-alphago-how-it-works

割鷄焉用牛刀 ( 할계언용우도 )

기업내사용자별분석요구 현업사용자 애널리스트 데이터사이언티스트 97 % ~ 3 % < 1 % 리포팅 Self-Service BI 원인분석 Forecasting 원인분석 Forecasting Custom Algorithm 개발

고급분석적용의변화 Predictive Modeling Predictive Modeling 데이터요청 업무지식전달 IT Data Scientists 데이터전달 Marketing Sales Manufacturing Planning SCM 업무지식전달 고급분석요청 분석결과전달 지원 지원 Marketing Sales Manufacturing Planning SCM Data Scientists IT

다양한분석요구를만족시키는 SAP Analytics Modules 현업사용자 애널리스트 데이터사이언티스트 97 % ~ 3 % < 1 % SAP Predictive Analytics HANA 플랫폼기능 : Predictive Analysis Library, R Vora on Spark SAP Business Intelligence Suite

SAP Analytics Modules SAP BusinessObjects BI Platform SAP Predictive Analysis SAP HANA Platform R Embedded R Predictive Analysis Library Text Processor Automated Predictive Library Data Service Text Data Processing DQ Transforms Data Pre-Processing and Loading SAP Data Services, Information Composer, SLT, DXC, Hadoop 2015 SAP SE or an SAP affiliate company. All rights reserved. 11

SAP Predictive Analytics SAP BusinessObjects BI Platform SAP Predictive Analysis SAP HANA Platform R Embedded R Predictive Analysis Library Text Processor Automated Predictive Library Data Service Text Data Processing DQ Transforms Data Pre-Processing and Loading SAP Data Services, Information Composer, SLT, DXC, Hadoop 2015 SAP SE or an SAP affiliate company. All rights reserved. 12

Predictive Analytics 현업사용자와데이터젂문가모두가편리하게홗용핛수있는단순화 자동화된데이터마이닝솔루션 SAP Predictive Analytics 구성요소 Automated Analytics 현업사용자와분석가들을위핚데이터마이닝자동화툴 Data manager / Modeler / Model manager Expert Analytics 젂문가들이오픈소스 R 을기반으로직접알고리즘을설계핛수있는툴 R configuration / visualization

Predictive Modeling Process UI 단순화전처리자동화모델자동화해석단순화 SQL 모델생성 Predictive Modeling Phase Business Questions 데이터준비및변홖예측모델개발업무적용비즈니스리뷰및모델갱신 Other Solutions SAP Predictive Analytics 2015 SAP SE or an SAP affiliate company. All rights reserved. 15

Predictive Modeling Process Business Question

Predictive Analytics 를통한설비고장예측 설비고장에대해사전적대응을핛수있는합리적인방안은 < 분석목표 > 1. 고장에가장많은영향을미치는요인탐색 2. 모델을통핚고장확률예측 3. 실시간으로예측결과적용

Predictive Maintenance 적용예시 - 데이터 timestamp cylinder n umber customer job number grain scree ned current density anode space ratio 센서수집정보, n=539 chrome content 1991-01-03 F108 MODMAT 47201 40 103.22 100 band Target 1991-01-04 B7 MODMAT 47201 YES 40 103.87 100 noband 1991-01-04 T133 MASSEY 39039 YES 40 108.06 100 noband 1991-01-04 T218 MASSEY 38039 YES 40 103.87 100 noband 1991-01-08 X126 TVGUIDE 25503 YES 40 105 100 band 1991-01-09 X266 TVGUIDE 25503 YES 40 105 100 noband 1991-01-11 J34 KMART 37351 NO 40 106.67 100 noband 1991-01-11 X249 ROSES 35751 NO 40 106.67 100 noband 39 개설명변수 Target 2015 SAP SE or an SAP affiliate company. All rights reserved. 19

Predictive Maintenance 적용예시 모델링결과 Predictive Power (KI) 모델의정확도 : 68.51% 설명변수들이 68.51% 로 Target 을설명핚다는의미 Prediction Confidence (KR) 모델의신뢰성 : 88.44% 학습모델과평가모델의차이가 11.56% 라는의미 95% 이하이므로더많은자료의수집이필요함 Nb. Variables Kept 39 개설명변수중 19 개의설명변수는제거되고, 20 개의변수만남았다는의미 2015 SAP SE or an SAP affiliate company. All rights reserved. 21

Predictive Maintenance 적용예시 고장원인분석 Variables Contribution ink type 0.13 press speed 0.11 Timestamp_Y 0.09 press 0.08 varnish pct 0.06 humifity 0.06 합계 1 2015 SAP SE or an SAP affiliate company. All rights reserved. 22

Predictive Maintenance 적용예시 고장확률예측 설비로그발생 고장예측모델적용 timestamp cylinder number customer job number grain screened current density anode space ratio chrome content 1991-01-03 F108 MODMAT 47201 40 103.22 100 1991-01-04 B7 MODMAT 47201 YES 40 103.87 100 1991-01-04 T133 MASSEY 39039 YES 40 108.06 100 1991-01-04 T218 MASSEY 38039 YES 40 103.87 100 1991-01-08 X126 TVGUIDE 25503 YES 40 105 100 1991-01-09 X266 TVGUIDE 25503 YES 40 105 100 1991-01-11 J34 KMART 37351 NO 40 106.67 100 1991-01-11 X249 ROSES 35751 NO 40 106.67 100 Target proba_rr 고장유무 0.807198984 0.79719577 0.797189742 0.787188892 0.757188066 0.74611849 0.695654012 0.695445417 고장발생확률 2015 SAP SE or an SAP affiliate company. All rights reserved. 25

Predictive Maintenance 적용예시 실시간예측모형적용 1 설비센싱정보 timestamp 2 Monitoring cylinder number customer job number grain screened current density anode space ratio chrome content 1991-01-03 F108 MODMAT 47201 40 103.22 100 1991-01-04 B7 MODMAT 47201 YES 40 103.87 100 고장확률예측 Stored Procedure 설비 ID 고장발생확률 원인 _ Ink Type 원인 _Press Speed E0004 85% Uncoated 1,900 3 설비담당자 Alert 예측모델에의핚고장원인정보 2015 SAP SE or an SAP affiliate company. All rights reserved. 27

고객사례 : John Deere < 소개 > 국가 : 미국 사업 : 농업중장비제조 매출 : 380억달러 (2013년) < 예지정비목적 > 품질보증관련비용 (warranty cost) 젃감 신속핚 supply part planning 엔지니어링개선을위핚 root cause 분석

고객사례 : John Deere <Predictive Maintenance Architecture> Telematics Message Processing (TMP) ESP Stream Transformation Situation Detection Automated Response Continuous Intelligence SAP Predictive Maintenance HANA Physical Table Virtual Data Model Predictive Analytics PAL BFL R Business Object BI R Studio Analytics App (Extreme Apps)

예시화면

800+ Global References SAP Predictive Analytics 는다양핚산업분야의 800 여개의고객사에서사용되고있습니다.

감사합니다 이철 Predictive Analytics 010-2987-6186 chul.lee@sap.com