Howdy! I'm Smith Shah

Curriculum Experience

Lean Manufacturing

Best Flow Line Equipments

Over the past one year, the main issues at Best flow have been the ever-rising inventory levels, ineffective space utilization and poor adherence to delivery schedules and we had to solve those issues. These numbers are: 

a. Inventory $14 Million

b. On-time delivery 58%

c. Space Utilization 45%

Tools used for analysis in this project include IDEF0, IDEF3, VSM, Spaghetti diagram and Shop floor Value Heat map. After laying down the ASIS details, we used the 8 TOBE flow design principles along with 8 cell design steps to draw a future state VSM and a rough shop floor layout for the improved production flow. Recommendations proposed include:

• Switching to the Make-to-order flow paradigm, especially for the products with less customization and high sales volumes.

• Reducing the changeover times by using the SMED methodology.

• Reducing batch size and thus improving the EPEI run values.

• Establishing a Kanban-induced pull system. 

• Changing layout from a scattered structure to a U-shaped structure.

• Performed a Cost-benefit analysis of all the recommended improvements in the process flow and layout.

• A temporal analysis of the Cost and benefits shows that the cumulative benefits exceed the cumulative cost after 7-8 years.

• In the TOBE analysis, a more compact U-shaped cell has been recommended, which not only offers a higher space utilization but also eliminates many NVA activities, especially in the logistics department and as a result we achieved 22% efficiency and 28% production rate improvements resulting in reduced WIP and transportation.

Advanced Quality Control

Anomaly Detection in a Manufacturing Process

• This project calls for Phase I analysis on a set of data collected from a manufacturing process.

•  This is a manufacturing related dataset of total of 552 data records. Each row is a data record. Each data record contains 209 values and the sample size is 𝑛 = 1. The physical meaning of each value is omitted.

• Main Goal: Remove Out-of-control points and Data Reduction to set the base for Phase II analysis. 

•  Multivariate Data Set = Multivariate Control Chart

  • T2

  • MCUSUM

  • MEWMA

• Used PCA in conjunction with Scree and Pareto Plots.

• 4 Principle Components used, 80% Variance Explained.

• Procedure starts with T2

  • Apply T2 iteratively until in-control.

• Apply MCUSUM

  • Apply MCUSUM iteratively until in-control.

• Repeat Process until neither chart has any out of control points.

• T2 is sensitive to spike and large sustained mean shift and MCUSUM is sensitive to small sustained mean shifts hence used in harmony.

• In total, 118 points removed indicating 434 in-control points remain and hence 78.6% of data preserved.

• Model explains 80.1% of variance, and preserves a lot of data and hence is robust.

• The in-control model has a UCL of 16.02 for T2 and 6 for m-CUSUM which can then be used for the Phase II analysis.

Demand Forecasting

Arconic Fastening System

• We had to buid Two short-term and two-long term time series forecast models to predict the heavy truck market rate until 2023 and also a simple forecasting tool for internal forecast generation with an ability to react to market swings.

• To achieve most accurate forecasting we did data exploration, we made an excel sheet to understand data.

• For the correlation we used pivot tables which can help you to identify the correlation between different customer and the products.

• Then we tried to use linear regression but due to lack of seasonality and huge amount of residuals we dropped the idea.

• This dropping of linear regression was concreted when we tried the plotting of regression line on all item groups.

• Once the models failed we thought to look for other models.

• To incorporate these models we used R programming.

• The different models we thought about were Exponential Time Smoothing, ARIMA, MAPA, TBATS,Holts Winter Method, Theta and linear regression.

• As the time series is irregular and there was no seasonality in some of the models, we dropped MAPA and Holts Winter method.

• Using the remaining method, we forecasted our models for a set of training data to check for the error in forecasting and using AIC, BIC and forecast values we tried to find best fit of the model.

Inventory Control

VTI Industries

• We had to develop a tool to determine safety stock, reorder point and reorder quantity, which automatically updates itself based on the usage and maximize the truckload capacity for each order along with an optimum product mix.

• With the basic understanding of the given data-sheet, we started by determining the policy we could use for calculating the safety stock, reorder point and the reorder quantity.

• There were two possibilities a continuous review system or a periodic review system.

• We decided to go forward with a periodic review system, implement the (s, S) policy and reduce the cost for the company.

• We then calculated the maximum truckload and the optimum product mix for the pool of the products given to us in the data using excel and calculated the required values and prepared a Linear Programming model to obtain the optimal truckload.

• In short,developed a user form in excel to find the safety stock, reorder point, and reorder quantity which automatically updates itself based on the input values and usage.

• Key variables include: usage with varying lead time, storage and shipment capacities for safety stock, reorder point and reorder quantity, and truckload maximization.

Markov Chains

Ranking of NFL Teams

Used Markov chain transition Algorithm to rank 32 NFL teams based on the weightage of votes casted corresponding to the margin of victory with the losing team voting for the winning team and another case In which both the teams cast votes according to the number of goals scored against them.

Poisson Distribution

Applications-Using Matlab

• Generated 10^6 samples using a random number generator and using those samples as the interarrival times,generated a sample poisson process and computed the expected number of paths and compared it with the actual number of paths.

• Created a histogram and the probability mass function for the Poisson random variable using matlab.

• Provided the evidence that conditional arrivals are uniformly distributed on a given time interval and plotted the empirical cummulative distribution function of arrival times and compared it with uniform CDF.