Designing the User Interface and Product Elements for a predictive maintainence tool for industrial grade machinary for Lecida.
Lecida is a predictive maintainance tool that effectively determines potential faliures and system anomalies in wind turbines and agro machines.
Lecida optimizes the performance of industrial machines through a unique combination of machine learning, distributed systems, and cloud computing. Their predictive technology alleviates the machine maintenance bottleneck manufacturers face. Lecida collects a plethora of data from thousands of machines around the world. My team was challenged with the task of determining what pieces of data are necessary for the user and how to most intuitively display/convey the relavent data points.
How might we determine what pieces of data are necessary for the user using the lecida program and how to most intuitively display/convey the relavent data points.
How might we present the most important information while not eliminating critical details that are required by the end user.
Our client Lecida was founded by a team of engineers from Stanford, Berkeley and Oxford with deep expertise in machine learning, distributed systems and cloud computing. It builds digital assistants for workers who either maintain or optimize the performance or availability of industrial machines.
To grasp the scope of this project, we investigated various industries (i.e. Agriculture, Wind Energy, etc.) and the respective roles of all individuals involved in the maintenance process. Lecida's potential clients include wind turbines, crop sprayers, scissor lifts and any type of industrial machine that can cause massive monetary and efficiency damage when they fail to function.
Based upon this research, we decided upon the five key elements of Lecida’s data: Location, Machine Type, Time to Failure, Severity Index, and Criticality. The Severity Index refers to the significance of the particular failure. Criticality is an amalgamation of the cited criteria represented by a score between 0 (least critical) and 1 (very critical).
The focus of our ideation is how to display predictions so that users can instantly understand which machine, which part of it, and when is it are failing. The first way to organize predictions is as notifications that users can click in to view details.
The purpose of the prediction page is to show predictions as notifications so that users can view them at a glance and click into them. When users click into each individual prediction, they are led to a more detailed page that lays out information about specific machine such as machine ID, location, time, and its failures.
Different couples care about, know, seek out different things, and are willing to put in different amounts of effort at different points in their search journeys.
Images are hugely important to get users to next step of process (to click to profile page, to request appt, etc) .
The Spot Estimate serves as a “mental checklist” that educates the user, but the SE is not the right place to surface that educational information
Design Iteration and Testing
We refined the Home/Prediction page and split the notifications as viewed and unseen. We added a side filter bar so that users can efficiently sort and find predictions by severity, location, time range, or machine type. When users click into each prediction, prediction detail is divided into machine information, sensor data, prediction history, and machine maintenance history. These added tabs will show users any machine-related information that will help them understand performance trend.
Mid Fidelity Iterations
We created a machine inventory page works like a "favorite" or "saved" page. If users want to put specific machines on the watchlist as priority, they can find the information card on the machine page and pin it so it stays on top as other prediction happen and gets updated.
It was also during this stage when we realized the need for an archive for predictions and failures since predictions do not equate failures. Having a log can allow users to track and look back to specific failures and correspondent predictions.
Hi Fidelity Designs and the Final Product
In the final product, we polished the already designed pages and introduced a reporting system. The reporting system allows users to report inaccurate prediction and unpredicted failure so that the AI system can adapt to these errors and perfect its prediction accuracy.
We chose a darkened teal as the base hue of our overall color scheme to create a sense of authority for our client’s image, and used variations of this darkened teal to complete our color palette.
Hi Fidelity Iterations
We chose Karla for its fun, trendy aesthetic. But bold font interface of Karla made the program interface easy to navigate and welcoming for users.
The use of color is separated the system into three levels: top, mid, and base. Base colors are grayscale and form the structure of all UI elements. Mid level colors are used for secondary functions, buttons, or any information less important to the user. Top level colors were chosen to stand out starkly against the base colors, and denote major functions, toggle & active states, and user inputs like sliders.