1. Introduction
Resilience has become crucial for manufacturing organizations, especially in the light of crises
experienced in recent years, such as the COVID-19 pandemic, supply chain disruptions, rising
energy prices, and political conflicts [
1
]. Resilience is defined as the ability of a system,
organization, or individual to withstand and recover from disruptions, shocks, or adversity and to adapt
and grow stronger in the face of challenges [
2, 3
]. Assessing resilience refers to the process of
measuring the resilience of an organization by examining its capacity to withstand and recover
from disruptions, adapt to changing conditions, and maintain operability. Resilience assessment
is essential for companies in today’s dynamic environment [
4
]. It helps organizations to identify
vulnerabilities, mitigate risks, maintain business continuity, be regulatory compliant, gain a
competitive edge, and foster stakeholder trust [
5
]. Nonetheless, systematic resilience assessment
is not considered in actual organizational risk management [
6
]. State-of-the-art approaches
CEUR
Workshop
Proceedings
such as enterprise risk management [
7
] recognize the importance of resilience, but the practical
implementation of resilience assessment is still limited due to the fuzzy nature of the term
resilience. For measuring resilience, several resilience indices have been developed to assess
the resilience of countries, cities, and organizations, e.g., Global Resilience Index (FM Global) ,
City Resilience Index1, Corporate Resilience Index (Resilinc) focusing on supply chains, Cyber
Resilience Index (World Economic Forum), or the National Risk Index (US Federal Emergency
Management Agency). Measuring the broader resilience of manufacturing organizations in all
facets, i.e., business units, is beyond the scope of these indices. In this work, we present GRACE
– a model for graph-based organizational resilience assessment in manufacturing. Our approach
works with centrality measures to identify influential nodes means key performance indicators
(KPIs) of organizational business units, e.g., time to fill open positions in human resources.
This enables highlighting critical KPIs that play a crucial role in organization functionality.
Furthermore, GRACE uses resilience metrics on KPI, business unit, and organizational level
for providing numerical measures of the organization’s ability to withstand disruptions and
recover from failures. By modeling those neuralgic points of the organization in the form of a
graph, simulated disruption scenarios can be fired for analyzing the organization’s response to
random or targeted disruptions. This helps in identifying vulnerabilities and critical points that
could cause cascading disruptions. Furthermore, potential response strategies can be tested
with high explainability for end users. GRACE was exemplified within a simulation service for
risk and crisis management in manufacturing. We evaluated the proposed approach in a case
study with the developed prototype in terms of performance in assessing the resilience of a
manufacturing company for two simulated disruption scenarios.
2. Graph-based resilience assessment: Model and case study
Related work shows several contributions with predominantly qualitative approaches for
assessing resilience in manufacturing organizations [
8
], e.g., in supply chain management [
9
]. Still, an
open issue is the practical operational measurement of the fuzzy term resilience [
10, 11
].Graph
theory and network analysis have been applied to assessing the resilience in processes, ,
sociotechnical systems [
12
], or supply chains [
13
]. The latter aims at developing a model to quantify
supply chain resilience as a single numerical value - the resilience index, that measures the
resilience capability of a company’s supply chain. We present GRACE, a model for
organizational resilience assessment in manufacturing extending the aforementioned work on supply
chains by Agarwal et al. [
13
] in terms of multiple resilience metrics for capturing all business
units of organizations (cf. Figure 1). GRACE operates on a knowledge graph obtained from
organizational data of relevant KPIs of business units and potential disruption categories in the
domain of interest of an organization. KPIs and DisruptionScenarios represent Nodes within
an Undirected Graph. KPIs, as for instance, employee satisfaction, are operationalized by a
metric [0...1] and belong to a Scope, i.e., a business unit such as production, or human resources
(cf. Figure 1). With a ScopeIndex [0...1], Scopes also define a metric for assessing the resilience
of single business units. Those indices are aggregated to the ResilienceIndex [0...1], measuring
the resilience of the whole organization, including all business units. DisruptionScenarios
1https://www.cityresilienceindex.org/
are classes of disruptions that are learned from historical data of the organization, e.g., past
power outages, fluctuations in energy prices, and supplier failures. They are characterized
by a TimeDimension representing the time horizon a disruption has an influence on an
organization’s functionality, e.g., 1-3 hours, several weeks (cf. Figure 1). Based on historical
data, DisruptionScenarios define potential ActionCategories that can be applied in case of
such a disruption, e.g., increase inventories, expand supplier network. They are characterized
by a list of KPIs they load on, means KPIs that would be influenced positively or negatively
by the action (KPI-Influence ). Combined with the aggregated degree of those KPI nodes
(KPI-Correlation) and learned probabilities of action success, we can measure the loading
power of an action, means the impact of a conducted action on an organization’s functionality
(cf. Figure 1). Last, DisruptionScenarios have a BackgroundRisk that combines the learned
probability of occurrence of a DisruptionScenario with a weight given by its expected
TimeDimension. Actual disruptions or crisis events, e.g., regulatory changes with respect to per- and
polyfluorinated chemicals (PFAS), can be induced as DisruptionItems that are classified as
one of the known DisruptionScenarios and instantiated with an ActualImpact combining
the derived BackgroundRisk with a probability for the concrete event (cf. Figure 1). GRACE
implements a process of graph-based resilience assessment consisting of three steps: (1) system
modeling, (2) attributes assignment, and (3) disruption scenario evaluation. In system modeling,
the organization is modeled as a graph with nodes representing KPIs and DisruptionScenarios.
Edges between both types of nodes are created based on historical data on past disruption
events, applied actions, and changes in KPIs during this time. Last, KPIs are clustered in Scopes
(business units). Second, attributes are assigned to nodes and edges for quantifying their
characteristics and interdependencies. KPIs are operationalized by a metric, representing their
status quo, e.g., based on ERP data. This enables the deduction of the actual ScopeIndex as
well as the ResilienceIndex. For all DisruptionScenarios, attributes such as the TimeDimension,
BackgroundRisk, and potential actions are determined and assigned. The edges between KPIs
and DisruptionScenarios are characterized by the ActionCategories that can be applied in case of
disruptions and that have an influence on KPIs. Graph-based analyses are applied to determine
degree centrality for the concepts KPI-Influence and KPI-Correlation. Last, scenario-based
evaluations are performed to assess the organization’s resilience under potential disruptions.
This involves simulating specific events, such as epidemics, cyber-attacks, or natural disasters,
and analyzing their impact on the robustness and performance of business units and the overall
organization. Thus, concrete DisruptionItems are fired onto the graph, inducing changes in KPIs,
indices of scopes as well as the ResilienceIndex. Potential ActionCategories are proposed by
showing their impact on the resilience metrics and thus the organization’s ability to withstand
disturbances and recover from failures. In summary, potential weaknesses can be identified and
strategies to enhance resilience can be designed based on GRACE.
Based on the proposed GRACE model (cf. Figure 1), we implemented a service for risk and crisis
management in manufacturing in form of a web interface2. The service accepts descriptions
of Scopes and DisruptionScenarios as well as initial KPI values in JSON format. Based on this,
the organizations’ system is modeled as a graph, and attributes, e.g., resilience metrics, are
assigned to nodes and edges according to the aforementioned process of graph-based resilience
assessment. The graph is displayed, and users can select DisruptionItems for disruption scenario
evaluation. Impacts of the disruption as well as potential ActionCategories with respect to
KPIs, scopes, and resilience indices are shown within the graph as well as in table format. As a
preliminary validation of the proposed model, we conducted a case study with the implemented
service to evaluate its performance in assessing the organizational resilience for two simulated
disruption cases using data from German manufacturing organizations of the research project
SPAICER3. The first case simulated supply chain disruptions, e.g., closure of key logistics hubs
due to COVID-19, afecting KPIs as the dependence on suppliers and machines’ fault tolerance
in production. The second case simulated an influenza epidemic, impacting KPIs like employee
retention and time-to-fill open positions in human resources. In both cases, the service
implementing GRACE successfully identified and located decreases in the organizations’ resilience
and recommended appropriate actions to mitigate the impact. Detailed results, including the
changes in relevant KPIs and recommended actions can be found in our Git repository2.