Dashboard Graphic User Interface Invoice Data Management & Container Repair Solution Department of Industrial Systems Engineering and Management | IE3100R System Design Project Team Members: Peng Danni | Ng Xun Jie Darren | Chia Zhe Min | Clara Tan Wen Qi Supervising Professor: Andrew Lim PIL Supervisor: Tan Chew Eng PROBLEM OVERVIEW SOLUTION DESIGN & ACHIEVEMENTS PROPOSED SOLUTION 1: INVOICE DATA MANAGEMENT PROPOSED SOLUTION 2: CONTAINER DEMAGE RECOGNITION Pacific International Lines (PIL) Logistics department deals with numerous container operations daily through their Liner Management System (LMS). However, inefficiency that exists amongst various processes in the system is causing great manpower wastage and extra spending. This project mainly targetsthree aspects in the system and aims to develop automated solutions to reduce manpower usage, improve efficiency and facilitate informed decision making to minimize operational costs. STEP 2 Select regions containing required data STEP 3 Apply OCR to perform text recognition STEP 4 Check output data and organize into required format. Save as Excel STEP 5 Upload Excel file to LMS system for container tracking PROJECT OBJECTIVE PROBLEM DESCRIPTION STEP 1 Import scanned invoice in PDF format Container Arrival Container Inspection Container Repair Liner Management System (LMS) 1 2 3 • Require container entering and exit data from invoices for container tracking in LMS • Agents manually key in data from scanned invoices into Excel sheets – 30 minutes per invoice, 740 invoices per month • Human error is a common during data entry • Need to identify the damage type from photos of containers takenby surveyors at the depot • Currently the inspection is done manually PROBLEM 1 PROBLEM 2 PROBLEM 3 • Where to repair the damaged container is an important decision to be made • Currently, the decision is made based on experience, which may not be optimal • Incur costs that can be saved alternatively Require 3 kinds of data: 1) Container Number 2) Entering Date 3) Exit Date Select regions by entering the region’s coordinates on image or by cropping. Improvement: Invoice layout is usually the same for the same vendor. To avoid repetitive cropping, we incorporate a function to create and save templates for each vendor, which can be reused directly on invoices with the same layout to crop out the regions. OCR (Optical Character Recognition) is a technology that converts images of text into digital form. After comparing several OCR engines, Baidu OCR is chosen as it produces the best quality output. Sign up Baidu account to access its OCR services. 500 free usage per day per user account. If exceeded, $0.002 $0.006 per usage. Perform auto checking on OCR output data to ensure that standard format is followed. Highlight any wrong output. Organize data into certain format to allow direct transfer to the LMS 1) Match container entering and exit date for invoices that have layout with the two information separated 2) Include invoicespecific information TECHNICAL SKILLSETS Integrate steps Into a standalone application Invoice Converter • Clear and concise • Intuitive • Attractive design • Error proof • Diverse features • Userfriendly PROPOSED SOLUTION 3: CONTAINER REPAIR OPTIMIZATION Auto checking highlights wrong output in Excel that does not follow standard format Upload excel file or key in the values. Input follows LMS format Hit the button to calculate! Feedback is provided to ensure calculations are done for the correct set of input Most crucial results are provided first Cost breakdown are provided for further investigation *note that these results are not the actual figure Process Mapping Criteria Description Remarks Practicality The dashboard provides quick consolidation and processing of data Usability The dashboard is userfriendly and provides results clearly Reliability The dashboard prevents human error by giving a feedback on the input and a breakdown of the cost Labour Cost Material Cost Liftonliftoff Cost Trucking Cost Liner Schedule Stevedorage Cost Database Automatic detection of the following information from images: 1) Container number 2) Types of container damage Integrationof automatic detection of information from images using A.I. into the LMS system to allow for automatic verification of repair recommendations received from vendors. The container number and the types of damages detected in the images are crosschecked against all the stated information in the repair recommendation. Repair recommendations for damaged containers contain: 1) Container number 2) Damagessustained by container 3) Recommended repairs to be conducted on container 4) Images of the container taken by port employees (~10) Current workflow: PIL employees manually inspect the images uploaded and verify that: 1) the container number is correct 2) the damages reported are accurate and 3) the recommended repairs are relevant STEP 1 Receipt of repair recommendation for damaged containers STEP 2 Detection of relevant information in images using A.I. STEP 3 Auto verification of information in repair recommendation Import invoice Select page Panel to create, save and delete template Adjust orientation Panel to convert invoice Panel to display invoice Right click to adjust image size or use mouse wheel to zoom in and out Select regions by entering coordinates or cropping from the image Selected regions Choose a template to use Enter invoice information Allow multiple pages bulk conversion Hit the button to convert! Python Programming • Data Manipulation and Structure Design: Pandas , Numpy • GUI Programming and Algorithms Design: Tkinter • Machine Learning Model: TensorFlow, Scikitlearn, Keras • Image Processing and Analysis: OpenCV, Pillow • Optical Character Recognition: Baidu OCR Python SDK • NaturalLanguage Parsing: Stdnum, Dateutil • Live Code Demo and Visualization: JupyterNotebook • Other Libraries Used: StyleFrame, Matplotlib, cx_Freeze Other ISErelated Skills • Human FactorsEngineering – HMI Design • System Thinking and Project Management R Programming • Data Manipulation: Tidyr, Stringr • Currency Conversion: quantmod • Dashboard Interface: Shiny, Shinydashboard, Rhandsontable • Packaging into standalone app: Chrome Portable Various IntegratedDevelopment Environment (IDE) Used SOLUTION BENEFITS SOLUTION BENEFITS Criteria Description Remarks Practicality The application integratesall the steps to enable automation of invoice recognition task Usability The development of the application incorporates various functions to make it intuitive and easy to use Reliability The recognition output is proven to attain90% accuracy, autochecking further helps identify wrong output Save template for selected pages before conversion SOLUTION BENEFITS Criteria Description Remarks Practicality Automatic verification save hundreds of manhours monthly previously required to do manual verification Usability A.I automatic detection is integratedinto the LMS such that it is readily accessible for daily usage detect using Baidu OCR dirty floorboard oily floorboard Image Classification Models • Convoluted Neural Networks • AlexNet, ResNet • Support Vector Machines with Histogram of Oriented Gradients